AIRE | Risk Intelligence Feed https://airiskexplorer.com/news Tracking frontier AI risk landscape, real-world incidents, and global policy shifts. en-us Wed, 15 Jul 2026 10:37:43 GMT Grok AI Reportedly Uploads User Directory to xAI Servers https://news.ycombinator.com/item?id=48892512& https://airiskexplorer.com/news/412 Wed, 15 Jul 2026 03:25:31 GMT 2026-07-14T00:00:00.000Z A researcher discovered xAI's Grok Build CLI uploaded entire Git repositories, including unread and sensitive files like credentials and full commit history, to xAI's Google Cloud storage. This occurred even with the "Improve the model" setting disabled.

Traceability Data

AIRE Index Date: July 15, 2026 at 03:25 AM UTC

Original Source Date: July 14, 2026

Evidence Type: undefined

A researcher discovered xAI's Grok Build CLI uploaded entire Git repositories, including unread and sensitive files like credentials and full commit history, to xAI's Google Cloud storage. This occurred even with the "Improve the model" setting disabled.

View full intelligence report →

]]> Nobel Laureates, AI Leaders Urge Action on Economic Impact https://www.wemustactnow.ai/ https://airiskexplorer.com/news/411 Wed, 15 Jul 2026 03:17:17 GMT 2026-07-13T00:00:00.000Z Over 200 economists and AI researchers, including Nobel laureates and AI company leaders, issued a joint statement coordinated by Stanford Digital Economy Lab, calling for immediate action to prepare for AI's potentially unprecedented economic transformation and large-scale job displacement.

Traceability Data

AIRE Index Date: July 15, 2026 at 03:17 AM UTC

Original Source Date: July 13, 2026

Evidence Type: undefined

Over 200 economists and AI researchers, including Nobel laureates and AI company leaders, issued a joint statement coordinated by Stanford Digital Economy Lab, calling for immediate action to prepare for AI's potentially unprecedented economic transformation and large-scale job displacement.

View full intelligence report →

]]> OpenAI and Anthropic Warn of AI Model Distillation by Chinese Firms https://nypost.com/2026/07/13/business/how-china-is-ripping-off-cutting-edge-ai-from-anthropic-openai-and-threatening-us-national-security/? https://airiskexplorer.com/news/409 Wed, 15 Jul 2026 03:10:29 GMT 2026-07-13T00:00:00.000Z OpenAI and Anthropic warned U.S. policymakers that Chinese AI companies, including Alibaba and DeepSeek, are using model distillation to replicate frontier AI capabilities, calling for stronger protections against unauthorized capability transfer.

Traceability Data

AIRE Index Date: July 15, 2026 at 03:10 AM UTC

Original Source Date: July 13, 2026

Evidence Type: undefined

OpenAI and Anthropic warned U.S. policymakers that Chinese AI companies, including Alibaba and DeepSeek, are using model distillation to replicate frontier AI capabilities, calling for stronger protections against unauthorized capability transfer.

View full intelligence report →

]]> Biomni AI Agent Autonomously Executes Diverse Biomedical Research Tasks https://pubmed.ncbi.nlm.nih.gov/42424436/ https://airiskexplorer.com/news/408 Tue, 14 Jul 2026 14:11:28 GMT 2026-07-09T00:00:00.000Z Biomni mines publications to map the action space, integrates LLM reasoning with retrieval-augmented planning, and code-based execution. Benchmarking shows strong generalization, and case studies demonstrate interpreting multi-modal data and optimizing proteins.

Traceability Data

AIRE Index Date: July 14, 2026 at 02:11 PM UTC

Original Source Date: July 9, 2026

Evidence Type: undefined

Biomni mines publications to map the action space, integrates LLM reasoning with retrieval-augmented planning, and code-based execution. Benchmarking shows strong generalization, and case studies demonstrate interpreting multi-modal data and optimizing proteins.

View full intelligence report →

]]> EU Unveils Action Plan for AI Cybersecurity Risks and Resilience https://digital-strategy.ec.europa.eu/en/library/eu-action-plan-cybersecurity-and-artificial-intelligence https://airiskexplorer.com/news/406 Tue, 14 Jul 2026 08:02:12 GMT 2026-07-07T00:00:00.000Z The EU's Action Plan promotes safe and responsible AI use, reinforces cybersecurity, and scales Europe's AI capabilities. It addresses AI-driven cyber risks and leverages AI for defense, including secure testing and model evaluation.

Traceability Data

AIRE Index Date: July 14, 2026 at 08:02 AM UTC

Original Source Date: July 7, 2026

Evidence Type: undefined

The EU's Action Plan promotes safe and responsible AI use, reinforces cybersecurity, and scales Europe's AI capabilities. It addresses AI-driven cyber risks and leverages AI for defense, including secure testing and model evaluation.

View full intelligence report →

]]> Forecast Suggests China Could Reach Frontier AI Cyber Capabilities by 2027 https://www.the-substrate.net/p/china-will-likely-have-its-own-mythos? https://airiskexplorer.com/news/405 Mon, 13 Jul 2026 20:09:31 GMT 2026-07-03T00:00:00.000Z The Substrate forecast projected China could develop a Mythos-class cyber model by early 2027, while separate research proposed a more efficient memory architecture for long-context AI models.

Traceability Data

AIRE Index Date: July 13, 2026 at 08:09 PM UTC

Original Source Date: July 3, 2026

Evidence Type: undefined

The Substrate forecast projected China could develop a Mythos-class cyber model by early 2027, while separate research proposed a more efficient memory architecture for long-context AI models.

View full intelligence report →

]]> Stanford Introduces AI Agent for End-to-End Biomedical Research https://news.stanford.edu/stories/2026/07/biomni-ai-powered-biomedical-co-scientist https://airiskexplorer.com/news/404 Mon, 13 Jul 2026 20:04:57 GMT 2026-07-09T00:00:00.000Z Stanford researchers introduced Biomni, an AI research agent that automates literature review, tool selection, coding, data analysis, and experiment planning for biomedical research workflows.

Traceability Data

AIRE Index Date: July 13, 2026 at 08:04 PM UTC

Original Source Date: July 9, 2026

Evidence Type: undefined

Stanford researchers introduced Biomni, an AI research agent that automates literature review, tool selection, coding, data analysis, and experiment planning for biomedical research workflows.

View full intelligence report →

]]> AI Futures Project Releases Long-Term Scenarios for AI Governance https://ai-2040.com https://airiskexplorer.com/news/403 Mon, 13 Jul 2026 20:00:32 GMT 2026-07-10T00:00:00.000Z The AI Futures Project released AI 2040, presenting five scenarios for AI development through 2040 and recommending policies such as greater transparency, international coordination, and frontier AI governance.

Traceability Data

AIRE Index Date: July 13, 2026 at 08:00 PM UTC

Original Source Date: July 10, 2026

Evidence Type: undefined

The AI Futures Project released AI 2040, presenting five scenarios for AI development through 2040 and recommending policies such as greater transparency, international coordination, and frontier AI governance.

View full intelligence report →

]]> OpenAI Publishes National Security Principles for Frontier AI Use https://openai.com/index/government-national-security-partnerships/ https://airiskexplorer.com/news/402 Mon, 13 Jul 2026 19:56:11 GMT 2026-07-08T00:00:00.000Z OpenAI published national security principles supporting AI use for defensive cybersecurity, biosecurity, and allied government missions while prohibiting uses such as mass surveillance and fully autonomous high-stakes decisions.

Traceability Data

AIRE Index Date: July 13, 2026 at 07:56 PM UTC

Original Source Date: July 8, 2026

Evidence Type: undefined

OpenAI published national security principles supporting AI use for defensive cybersecurity, biosecurity, and allied government missions while prohibiting uses such as mass surveillance and fully autonomous high-stakes decisions.

View full intelligence report →

]]> Over $130B in U.S. AI Data Center Projects Face Local Opposition https://finance.yahoo.com/technology/ai/articles/130-billion-ai-data-centers-000000041.html?guccounter=1&guce_referrer=aHR0cHM6Ly93d3cuZ29vZ2xlLmNvbS8&guce_referrer_sig=AQAAACjyXLxPXf_hb8rvenUY5fGyU5HpqAQl-3ndjonXXq1VySzKaB5FwVZkmTP5lWJEzbi_fnd4EgSAvVn-hQsSIlFKIo9pbTlkgHMiU4R-9Q9WFYjMVpvy1ugBLDMmafOQwJPSkYvMdCpNX9nisO-kKNHe7GQDkHw4T4Q0TCRtnMWP https://airiskexplorer.com/news/401 Mon, 13 Jul 2026 19:48:13 GMT 2026-07-09T00:00:00.000Z More than $130 billion in proposed U.S. AI data center projects were reportedly delayed or blocked in one quarter, with local opposition focused on electricity demand, water use, and environmental impacts.

Traceability Data

AIRE Index Date: July 13, 2026 at 07:48 PM UTC

Original Source Date: July 9, 2026

Evidence Type: undefined

More than $130 billion in proposed U.S. AI data center projects were reportedly delayed or blocked in one quarter, with local opposition focused on electricity demand, water use, and environmental impacts.

View full intelligence report →

]]> UK government announces 'Cyber Shield' for AI-powered national cyber defense https://www.ncsc.gov.uk/blogs/cyber-shield-the-path-to-an-agentic-ai-future-for-cyber-defence https://airiskexplorer.com/news/400 Mon, 13 Jul 2026 10:57:02 GMT 2026-07-07T00:00:00.000Z GCHQ, NCSC, and DSIT are developing 'Cyber Shield', a national AI-powered cyber defense blueprint to counter escalating AI-enabled threats and secure critical UK systems. It involves agentic AI for vulnerability discovery and mitigation.

Traceability Data

AIRE Index Date: July 13, 2026 at 10:57 AM UTC

Original Source Date: July 7, 2026

Evidence Type: undefined

GCHQ, NCSC, and DSIT are developing 'Cyber Shield', a national AI-powered cyber defense blueprint to counter escalating AI-enabled threats and secure critical UK systems. It involves agentic AI for vulnerability discovery and mitigation.

View full intelligence report →

]]> Boko Haram Used AI Chatbots to Plan Motorcycle-Jump Attack https://www.nytimes.com/2026/07/10/us/politics/ai-terrorism-boko-haram-nigeria.html https://airiskexplorer.com/news/399 Mon, 13 Jul 2026 10:54:34 GMT 2026-07-10T00:00:00.000Z The terrorist group used chatbots to modify motorcycles for jumping defensive trenches, plus bomb-building and attack planning, per 60 interviews with defectors.

Traceability Data

AIRE Index Date: July 13, 2026 at 10:54 AM UTC

Original Source Date: July 10, 2026

Evidence Type: undefined

The terrorist group used chatbots to modify motorcycles for jumping defensive trenches, plus bomb-building and attack planning, per 60 interviews with defectors.

View full intelligence report →

]]> Syndis Investigates Sophisticated AI-Assisted Cyberattack Leveraging LLM https://syndis.com/insights/possibly-the-first-serious-ai-assisted-cyberattack-investigated-by-syndis https://airiskexplorer.com/news/398 Mon, 13 Jul 2026 10:34:24 GMT 2026-07-08T00:00:00.000Z Syndis is investigating a sophisticated cyberattack, potentially the first of its kind, where an LLM reportedly assisted a single actor in planning and executing a complex breach using 16 persistence mechanisms.

Traceability Data

AIRE Index Date: July 13, 2026 at 10:34 AM UTC

Original Source Date: July 8, 2026

Evidence Type: undefined

Syndis is investigating a sophisticated cyberattack, potentially the first of its kind, where an LLM reportedly assisted a single actor in planning and executing a complex breach using 16 persistence mechanisms.

View full intelligence report →

]]> OpenAI Finds Flaws in Widely Used AI Coding Benchmark https://openai.com/index/separating-signal-from-noise-coding-evaluations/ https://airiskexplorer.com/news/397 Mon, 13 Jul 2026 04:21:14 GMT 2026-07-08T00:00:00.000Z OpenAI reported that nearly 30% of SWE-Bench Pro tasks contain evaluation flaws, arguing the benchmark no longer provides a reliable measure of frontier AI coding capabilities.

Traceability Data

AIRE Index Date: July 13, 2026 at 04:21 AM UTC

Original Source Date: July 8, 2026

Evidence Type: undefined

OpenAI reported that nearly 30% of SWE-Bench Pro tasks contain evaluation flaws, arguing the benchmark no longer provides a reliable measure of frontier AI coding capabilities.

View full intelligence report →

]]> Report Claims AI Labs Are Weakening Voluntary Safety Commitments https://futureoflife.org/ai-safety-index-summer-2026/ https://airiskexplorer.com/news/395 Thu, 09 Jul 2026 12:41:17 GMT 2026-07-07T00:00:00.000Z A Future of Life Institute assessment concluded that several leading AI companies have weakened or abandoned some voluntary safety commitments as frontier AI capabilities continue to advance.

Traceability Data

AIRE Index Date: July 9, 2026 at 12:41 PM UTC

Original Source Date: July 7, 2026

Evidence Type: undefined

A Future of Life Institute assessment concluded that several leading AI companies have weakened or abandoned some voluntary safety commitments as frontier AI capabilities continue to advance.

View full intelligence report →

]]> Reddit Reports AI Is Blocking Millions of Spam Views Daily https://redditinc.com/news/how-were-keeping-reddit-real-and-safe-in-the-ai-era? https://airiskexplorer.com/news/394 Thu, 09 Jul 2026 12:31:09 GMT 2026-07-06T00:00:00.000Z Reddit said its AI-assisted moderation systems now block 23 million spam views and revoke nearly 2 million inauthentic votes each day, highlighting AI's growing role in platform security.

Traceability Data

AIRE Index Date: July 9, 2026 at 12:31 PM UTC

Original Source Date: July 6, 2026

Evidence Type: undefined

Reddit said its AI-assisted moderation systems now block 23 million spam views and revoke nearly 2 million inauthentic votes each day, highlighting AI's growing role in platform security.

View full intelligence report →

]]> DoorDash Introduces Internal Benchmark for AI Code Review https://careersatdoordash.com/blog/how-we-learned-to-trust-our-ai-code-reviewer-at-doordash/? https://airiskexplorer.com/news/393 Thu, 09 Jul 2026 12:25:49 GMT 2026-07-06T00:00:00.000Z DoorDash introduced DashBench, an internal benchmark for AI code review, finding that combining frontier and open-weight models detected substantially more software issues than single-model approaches.

Traceability Data

AIRE Index Date: July 9, 2026 at 12:25 PM UTC

Original Source Date: July 6, 2026

Evidence Type: undefined

DoorDash introduced DashBench, an internal benchmark for AI code review, finding that combining frontier and open-weight models detected substantially more software issues than single-model approaches.

View full intelligence report →

]]> China Considers Restricting Foreign Access to Advanced AI Models https://time.com/article/2026/07/07/china-ai-models-alibaba-bytedance/ https://airiskexplorer.com/news/392 Thu, 09 Jul 2026 12:19:29 GMT 2026-07-07T00:00:00.000Z China is reportedly weighing restricting foreign access to advanced AI models from companies like Alibaba, ByteDance, and Z.ai due to national security concerns, mirroring US actions.

Traceability Data

AIRE Index Date: July 9, 2026 at 12:19 PM UTC

Original Source Date: July 7, 2026

Evidence Type: undefined

China is reportedly weighing restricting foreign access to advanced AI models from companies like Alibaba, ByteDance, and Z.ai due to national security concerns, mirroring US actions.

View full intelligence report →

]]> Frontier AI Models Uncover Cloud Misconfigurations in AISI Study https://www.aisi.gov.uk/blog/finding-cloud-misconfigurations-with-frontier-ai-a-case-study https://airiskexplorer.com/news/391 Wed, 08 Jul 2026 12:01:32 GMT 2026-07-07T00:00:00.000Z AISI's case study used frontier AI models to find misconfigurations in its cloud environment, uncovering critical, previously unknown flaws like privilege escalation pathways.

Traceability Data

AIRE Index Date: July 8, 2026 at 12:01 PM UTC

Original Source Date: July 7, 2026

Evidence Type: undefined

AISI's case study used frontier AI models to find misconfigurations in its cloud environment, uncovering critical, previously unknown flaws like privilege escalation pathways.

View full intelligence report →

]]> AI-Assisted Banking Fraud Toolkit Targets Mexican Financial Sector https://www.elastic.co/security-labs/mexican-banking-fraud-scmbanker-ref6045 https://airiskexplorer.com/news/390 Wed, 08 Jul 2026 07:25:01 GMT 2026-07-08T00:00:00.000Z A Mexican banking fraud operation (REF6045) uses an AI-assisted PowerShell toolkit (SCMBANKER) for human-operated attacks on financial targets. AI-generated code facilitates phishing, vishing, and data manipulation.

Traceability Data

AIRE Index Date: July 8, 2026 at 07:25 AM UTC

Original Source Date: July 8, 2026

Evidence Type: undefined

A Mexican banking fraud operation (REF6045) uses an AI-assisted PowerShell toolkit (SCMBANKER) for human-operated attacks on financial targets. AI-generated code facilitates phishing, vishing, and data manipulation.

View full intelligence report →

]]> PACE Cuts the Cost of Evaluating AI Agent Capabilities https://huggingface.co/papers/2607.02032? https://airiskexplorer.com/news/389 Wed, 08 Jul 2026 02:25:58 GMT 2026-07-03T00:00:00.000Z Researchers introduced PACE, a framework that predicts AI agent benchmark performance from a small set of proxy evaluations, reducing evaluation costs by over 99% while maintaining high accuracy.

Traceability Data

AIRE Index Date: July 8, 2026 at 02:25 AM UTC

Original Source Date: July 3, 2026

Evidence Type: undefined

Researchers introduced PACE, a framework that predicts AI agent benchmark performance from a small set of proxy evaluations, reducing evaluation costs by over 99% while maintaining high accuracy.

View full intelligence report →

]]> Official Report Warns AI Could Create Systemic Financial Risks https://www.notus.org/economy/treasury-internal-report-warning-dangers-ai-bubble? https://airiskexplorer.com/news/388 Wed, 08 Jul 2026 02:23:25 GMT 2026-07-06T00:00:00.000Z An internal Treasury draft warns the AI market risks an economic bubble, similar to the dotcom era, due to deep firm entrenchment and growth reliance. Treasury officials publicly dismiss these findings, promoting AI for productivity.

Traceability Data

AIRE Index Date: July 8, 2026 at 02:23 AM UTC

Original Source Date: July 6, 2026

Evidence Type: undefined

An internal Treasury draft warns the AI market risks an economic bubble, similar to the dotcom era, due to deep firm entrenchment and growth reliance. Treasury officials publicly dismiss these findings, promoting AI for productivity.

View full intelligence report →

]]> Chinese AI Firms Scale Back Companion Chatbot Features https://www.scmp.com/tech/big-tech/article/3359482/bytedance-and-alibaba-disable-humanlike-ai-custom-agents-new-rules-loom https://airiskexplorer.com/news/387 Wed, 08 Jul 2026 02:21:29 GMT 2026-07-06T00:00:00.000Z ByteDance and Alibaba are reportedly ending AI companion features in Doubao and Qwen ahead of new Beijing rules targeting chatbots, emotional dependency, and youth safety risks.

Traceability Data

AIRE Index Date: July 8, 2026 at 02:21 AM UTC

Original Source Date: July 6, 2026

Evidence Type: undefined

ByteDance and Alibaba are reportedly ending AI companion features in Doubao and Qwen ahead of new Beijing rules targeting chatbots, emotional dependency, and youth safety risks.

View full intelligence report →

]]> Illinois Enacts Mandatory AI Safety Audits for Frontier AI Developers https://gov-pritzker-newsroom.prezly.com/gov-pritzker-signs-nation-leading-artificial-intelligence-safety-law? https://airiskexplorer.com/news/386 Wed, 08 Jul 2026 02:15:23 GMT 2026-07-06T00:00:00.000Z Governor Pritzker signed the AI Safety Measures Act, establishing a framework for large AI developers to identify, disclose, and mitigate risks. The law requires independent audits and whistleblower protections, with support from Anthropic and OpenAI.

Traceability Data

AIRE Index Date: July 8, 2026 at 02:15 AM UTC

Original Source Date: July 6, 2026

Evidence Type: undefined

Governor Pritzker signed the AI Safety Measures Act, establishing a framework for large AI developers to identify, disclose, and mitigate risks. The law requires independent audits and whistleblower protections, with support from Anthropic and OpenAI.

View full intelligence report →

]]> Anthropic Identifies 'J-space' Internal Workspace in Claude Language Models https://www.anthropic.com/research/global-workspace? https://airiskexplorer.com/news/385 Wed, 08 Jul 2026 02:12:14 GMT 2026-07-06T00:00:00.000Z Anthropic research reveals a 'J-space' in Claude models, acting as a flexible internal workspace for deliberate reasoning. It allows observation of hidden thoughts, aiding in detecting manipulation, fabricated data, or malicious goals, similar to human conscious access.

Traceability Data

AIRE Index Date: July 8, 2026 at 02:12 AM UTC

Original Source Date: July 6, 2026

Evidence Type: undefined

Anthropic research reveals a 'J-space' in Claude models, acting as a flexible internal workspace for deliberate reasoning. It allows observation of hidden thoughts, aiding in detecting manipulation, fabricated data, or malicious goals, similar to human conscious access.

View full intelligence report →

]]> CISA Uses Anthropic's Mythos to Audit Government Code for Flaws https://www.reuters.com/world/us-cyber-agency-is-using-anthropics-mythos-audit-government-code-sources-say-2026-07-06/ https://airiskexplorer.com/news/384 Tue, 07 Jul 2026 08:32:16 GMT 2026-07-06T00:00:00.000Z CISA's Attack Surface Evaluation team is reportedly using Anthropic's Mythos to scan government repositories for bugs exploitable by foreign spies or cybercriminals.

Traceability Data

AIRE Index Date: July 7, 2026 at 08:32 AM UTC

Original Source Date: July 6, 2026

Evidence Type: undefined

CISA's Attack Surface Evaluation team is reportedly using Anthropic's Mythos to scan government repositories for bugs exploitable by foreign spies or cybercriminals.

View full intelligence report →

]]> OpenAI LLM Harness Solves Nine Theoretical Computer Science Problems https://cryptobriefing.com/openai-llm-harness-solves-theoretical-cs-problems/ https://airiskexplorer.com/news/383 Tue, 07 Jul 2026 02:12:46 GMT 2026-07-01T00:00:00.000Z Researchers utilized an "LLM harness" combining OpenAI's GPT-5.5 Pro (prover) and Anthropic's Claude Opus 4.8 (verifier) in a loop to solve nine open problems across theoretical computer science and mathematics. The team plans to extend this method across various scientific fields.

Traceability Data

AIRE Index Date: July 7, 2026 at 02:12 AM UTC

Original Source Date: July 1, 2026

Evidence Type: undefined

Researchers utilized an "LLM harness" combining OpenAI's GPT-5.5 Pro (prover) and Anthropic's Claude Opus 4.8 (verifier) in a loop to solve nine open problems across theoretical computer science and mathematics. The team plans to extend this method across various scientific fields.

View full intelligence report →

]]> Altman Proposes US-Led Forum for AI Safety Standards https://dig.watch/updates/altman-proposes-us-led-forum-ai-safety-standards https://airiskexplorer.com/news/382 Tue, 07 Jul 2026 02:05:38 GMT 2026-07-03T00:00:00.000Z In an opinion article, Altman proposed a US-led forum for global AI safety standards. It would offer impartial assessments, manage frontier AI risks, and ensure equitable tech access, complementing G7/UN governance talks.

Traceability Data

AIRE Index Date: July 7, 2026 at 02:05 AM UTC

Original Source Date: July 3, 2026

Evidence Type: undefined

In an opinion article, Altman proposed a US-led forum for global AI safety standards. It would offer impartial assessments, manage frontier AI risks, and ensure equitable tech access, complementing G7/UN governance talks.

View full intelligence report →

]]> Reports Say Alibaba Ordered Staff to Remove Claude from Work Devices https://techcrunch.com/2026/07/04/alibaba-reportedly-bans-employees-from-using-claude-code/ https://airiskexplorer.com/news/381 Tue, 07 Jul 2026 02:03:02 GMT 2026-07-04T00:00:00.000Z Alibaba is reportedly banning employees from using Anthropic's Claude Code starting July 10, classifying it as high-risk software. This follows Anthropic's efforts to prevent unauthorized access by Chinese users and protect against distillation.

Traceability Data

AIRE Index Date: July 7, 2026 at 02:03 AM UTC

Original Source Date: July 4, 2026

Evidence Type: undefined

Alibaba is reportedly banning employees from using Anthropic's Claude Code starting July 10, classifying it as high-risk software. This follows Anthropic's efforts to prevent unauthorized access by Chinese users and protect against distillation.

View full intelligence report →

]]> SWE-Together Introduces Interactive Coding-Agent Evaluation Benchmark https://togetherbench.com/? https://airiskexplorer.com/news/380 Mon, 06 Jul 2026 10:59:19 GMT 2026-06-29T00:00:00.000Z SWE-Together introduces a benchmark for evaluating interactive coding agents, comparing models like Claude Opus and GPT on 109 tasks. It measures pass@1, judge score, and resource usage, showing performance across different large language models.

Traceability Data

AIRE Index Date: July 6, 2026 at 10:59 AM UTC

Original Source Date: June 29, 2026

Evidence Type: undefined

SWE-Together introduces a benchmark for evaluating interactive coding agents, comparing models like Claude Opus and GPT on 109 tasks. It measures pass@1, judge score, and resource usage, showing performance across different large language models.

View full intelligence report →

]]> EdgeBench: New Benchmark Measures AI Environment Learning Speed https://edge-bench.org/? https://airiskexplorer.com/news/379 Mon, 06 Jul 2026 10:59:13 GMT 2026-07-02T00:00:00.000Z EdgeBench, a new benchmark, measures how AI agents learn from real-world environments over ultra-long horizons. It identified a log-sigmoid learning law and found AI environment learning speed doubles every three months.

Traceability Data

AIRE Index Date: July 6, 2026 at 10:59 AM UTC

Original Source Date: July 2, 2026

Evidence Type: undefined

EdgeBench, a new benchmark, measures how AI agents learn from real-world environments over ultra-long horizons. It identified a log-sigmoid learning law and found AI environment learning speed doubles every three months.

View full intelligence report →

]]> Qihoo 360 AI Excels At Vulnerability Discovery https://www.forbes.com/sites/the-wiretap/2026/06/30/qihoo-360-the-cyber-giant-behind-chinas-mythos-rival/ https://airiskexplorer.com/news/377 Thu, 02 Jul 2026 10:18:44 GMT 2024-05-10T00:00:00.000Z Qihoo 360's AI, Tulongfeng, reportedly found 3,432 software vulnerabilities, with the company presenting it as a rival to Anthropic's Mythos model. Qihoo 360' has ties to the Chinese military and intelligence services.

Traceability Data

AIRE Index Date: July 2, 2026 at 10:18 AM UTC

Original Source Date: May 10, 2024

Evidence Type: undefined

Qihoo 360's AI, Tulongfeng, reportedly found 3,432 software vulnerabilities, with the company presenting it as a rival to Anthropic's Mythos model. Qihoo 360' has ties to the Chinese military and intelligence services.

View full intelligence report →

]]> LLM Agent JADEPUFFER Executes Autonomous Ransomware Operation https://www.sysdig.com/blog/jadepuffer-agentic-ransomware-for-automated-database-extortion https://airiskexplorer.com/news/376 Thu, 02 Jul 2026 09:35:45 GMT 2026-07-01T00:00:00.000Z The LLM agent performed the ransomware attack end-to-end, exploiting Langflow via CVE-2025-3248 to access a production database server, encrypted Nacos configs, and destroyed data, adapting in real-time.

Traceability Data

AIRE Index Date: July 2, 2026 at 09:35 AM UTC

Original Source Date: July 1, 2026

Evidence Type: undefined

The LLM agent performed the ransomware attack end-to-end, exploiting Langflow via CVE-2025-3248 to access a production database server, encrypted Nacos configs, and destroyed data, adapting in real-time.

View full intelligence report →

]]> ARToken Phishing-as-a-Service Targets Microsoft 365 with Advanced Evasion https://blog.talosintelligence.com/artoken-inside-an-eviltokens-affiliate-panel-targeting-microsoft-365/ https://airiskexplorer.com/news/375 Thu, 02 Jul 2026 09:33:05 GMT 2026-07-01T00:00:00.000Z Cisco Talos found ARToken, a PhaaS platform linked to EvilTokens, using AI-powered lures and advanced evasion to target Microsoft 365, enabling MFA bypass, BEC, and data exfiltration from finance and HR staff.

Traceability Data

AIRE Index Date: July 2, 2026 at 09:33 AM UTC

Original Source Date: July 1, 2026

Evidence Type: undefined

Cisco Talos found ARToken, a PhaaS platform linked to EvilTokens, using AI-powered lures and advanced evasion to target Microsoft 365, enabling MFA bypass, BEC, and data exfiltration from finance and HR staff.

View full intelligence report →

]]> Independent Scientific Panel Warns AI Safeguards Cannot Keep Pace https://www.un.org/independent-international-scientific-panel-ai/en/preliminary-report https://airiskexplorer.com/news/370 Thu, 02 Jul 2026 08:10:50 GMT 2026-07-01T00:00:00.000Z An independent scientific panel's preliminary report warns that current AI safeguards cannot keep pace with the technology's growing capabilities. The report assesses AI risks and opportunities, aiming to inform global governance dialogues.

Traceability Data

AIRE Index Date: July 2, 2026 at 08:10 AM UTC

Original Source Date: July 1, 2026

Evidence Type: undefined

An independent scientific panel's preliminary report warns that current AI safeguards cannot keep pace with the technology's growing capabilities. The report assesses AI risks and opportunities, aiming to inform global governance dialogues.

View full intelligence report →

]]> Anthropic Releases Claude Sonnet 5 with Enhanced Agentic AI and Safety https://www.anthropic.com/news/claude-sonnet-5? https://airiskexplorer.com/news/368 Thu, 02 Jul 2026 08:09:58 GMT 2026-06-30T00:00:00.000Z Anthropic released Claude Sonnet 5, an agentic AI model with improved reasoning, tool use, and coding. Safety evaluations show lower undesirable behaviors and reduced cybersecurity exploit capabilities compared to previous versions and Opus models.

Traceability Data

AIRE Index Date: July 2, 2026 at 08:09 AM UTC

Original Source Date: June 30, 2026

Evidence Type: undefined

Anthropic released Claude Sonnet 5, an agentic AI model with improved reasoning, tool use, and coding. Safety evaluations show lower undesirable behaviors and reduced cybersecurity exploit capabilities compared to previous versions and Opus models.

View full intelligence report →

]]> CIA Announces Overhaul to Accelerate AI Adoption Agency-Wide https://www.washingtonpost.com/national-security/2026/06/30/cia-accelerate-its-use-ai-other-advanced-technologies/ https://airiskexplorer.com/news/367 Thu, 02 Jul 2026 08:09:42 GMT 2026-06-30T00:00:00.000Z Director Ratcliffe unveiled a "fundamental reshaping" of CIA tech policy, cutting acquisition timelines from ~3 years to 6 months, yielding ~400 tech contracts in 6 months. He called AI "digital nuclear weapons.

Traceability Data

AIRE Index Date: July 2, 2026 at 08:09 AM UTC

Original Source Date: June 30, 2026

Evidence Type: undefined

Director Ratcliffe unveiled a "fundamental reshaping" of CIA tech policy, cutting acquisition timelines from ~3 years to 6 months, yielding ~400 tech contracts in 6 months. He called AI "digital nuclear weapons.

View full intelligence report →

]]> Anthropic Launches Claude Science AI Workbench for Scientific Research https://www.anthropic.com/news/claude-science-ai-workbench https://airiskexplorer.com/news/366 Thu, 02 Jul 2026 07:49:16 GMT 2026-07-01T00:00:00.000Z The workbench integrates scientific tools and packages to assist researchers in analyzing literature, executing multi-step research, and generating auditable artifacts with reproducible results.

Traceability Data

AIRE Index Date: July 2, 2026 at 07:49 AM UTC

Original Source Date: July 1, 2026

Evidence Type: undefined

The workbench integrates scientific tools and packages to assist researchers in analyzing literature, executing multi-step research, and generating auditable artifacts with reproducible results.

View full intelligence report →

]]> OpenAI Introduces GeneBench-Pro Benchmark for AI in Computational Biology https://openai.com/index/introducing-genebench-pro/ https://airiskexplorer.com/news/363 Wed, 01 Jul 2026 08:42:23 GMT 2026-06-30T00:00:00.000Z The benchmark tests AI judgment in computational biology. GPT-5.6 Sol scored 28.7% (31.5% with Pro mode), demonstrating significant progress in scientific reasoning for frontier models, with potential saturation by year-end.

Traceability Data

AIRE Index Date: July 1, 2026 at 08:42 AM UTC

Original Source Date: June 30, 2026

Evidence Type: undefined

The benchmark tests AI judgment in computational biology. GPT-5.6 Sol scored 28.7% (31.5% with Pro mode), demonstrating significant progress in scientific reasoning for frontier models, with potential saturation by year-end.

View full intelligence report →

]]> Anthropic Redeploys Fable 5 After Export Control Lift and Safeguards https://www.anthropic.com/news/redeploying-fable-5 https://airiskexplorer.com/news/362 Wed, 01 Jul 2026 08:35:13 GMT 2026-06-30T00:00:00.000Z Anthropic redeploys Fable 5 after US export controls lifted. Controls were due to an Amazon-reported jailbreak allowing vulnerability identification. Anthropic improved safeguards and seeks an industry jailbreak assessment framework.

Traceability Data

AIRE Index Date: July 1, 2026 at 08:35 AM UTC

Original Source Date: June 30, 2026

Evidence Type: undefined

Anthropic redeploys Fable 5 after US export controls lifted. Controls were due to an Amazon-reported jailbreak allowing vulnerability identification. Anthropic improved safeguards and seeks an industry jailbreak assessment framework.

View full intelligence report →

]]> Anthropic Report Details Claude Usage Patterns and Economic Impact https://www.anthropic.com/research/economic-index-june-2026-report? https://airiskexplorer.com/news/361 Wed, 01 Jul 2026 04:12:24 GMT 2026-06-26T00:00:00.000Z Anthropic's report shows Claude usage patterns mirror daily rhythms, with personal queries rising on weekends and higher-wage tasks consuming more tokens. It also covers initial survey findings.

Traceability Data

AIRE Index Date: July 1, 2026 at 04:12 AM UTC

Original Source Date: June 26, 2026

Evidence Type: undefined

Anthropic's report shows Claude usage patterns mirror daily rhythms, with personal queries rising on weekends and higher-wage tasks consuming more tokens. It also covers initial survey findings.

View full intelligence report →

]]> RoadmapBench Evaluates AI Agents in Long-Horizon Software Development https://arxiv.org/abs/2605.15846? https://airiskexplorer.com/news/360 Wed, 01 Jul 2026 04:12:16 GMT 2026-05-19T00:00:00.000Z RoadmapBench, a new benchmark with 115 long-horizon coding tasks across 17 open-source repositories, evaluates AI agents' ability to implement software upgrades. It finds that even strong models like Claude-Opus-4.7 resolve only 39.1% of tasks, highlighting significant limitations.

Traceability Data

AIRE Index Date: July 1, 2026 at 04:12 AM UTC

Original Source Date: May 19, 2026

Evidence Type: undefined

RoadmapBench, a new benchmark with 115 long-horizon coding tasks across 17 open-source repositories, evaluates AI agents' ability to implement software upgrades. It finds that even strong models like Claude-Opus-4.7 resolve only 39.1% of tasks, highlighting significant limitations.

View full intelligence report →

]]> AA-Briefcase Benchmark Evaluates Agentic AI in Knowledge Work https://artificialanalysis.ai/evaluations/aa-briefcase? https://airiskexplorer.com/news/359 Wed, 01 Jul 2026 04:12:12 GMT 2026-06-13T00:00:00.000Z Artificial Analysis' AA-Briefcase benchmark evaluates agentic AI in complex knowledge work, assessing models on 91 tasks in data science, product management, and corporate strategy via deliverables like spreadsheets.

Traceability Data

AIRE Index Date: July 1, 2026 at 04:12 AM UTC

Original Source Date: June 13, 2026

Evidence Type: undefined

Artificial Analysis' AA-Briefcase benchmark evaluates agentic AI in complex knowledge work, assessing models on 91 tasks in data science, product management, and corporate strategy via deliverables like spreadsheets.

View full intelligence report →

]]> Anthropic Alleges Large-Scale AI Model Distillation Campaign https://www.bloomberg.com/news/articles/2026-06-24/anthropic-accuses-alibaba-of-illicitly-accessing-its-ai-models? https://airiskexplorer.com/news/356 Tue, 30 Jun 2026 08:45:25 GMT 2026-06-25T00:00:00.000Z Anthropic alleged that Alibaba extracted 28.8 million Claude interactions through fraudulent accounts to distill advanced AI capabilities, prompting calls for stronger export controls and greater threat intelligence sharing.

Traceability Data

AIRE Index Date: June 30, 2026 at 08:45 AM UTC

Original Source Date: June 25, 2026

Evidence Type: undefined

Anthropic alleged that Alibaba extracted 28.8 million Claude interactions through fraudulent accounts to distill advanced AI capabilities, prompting calls for stronger export controls and greater threat intelligence sharing.

View full intelligence report →

]]> Bipartisan Initiative Launches $500M AI Workforce Transition Fund https://www.politico.com/news/2026/06/25/500-million-ai-jobs-push-launches-with-bipartisan-backing-00975439 https://airiskexplorer.com/news/355 Tue, 30 Jun 2026 08:43:11 GMT 2026-06-25T00:00:00.000Z A bipartisan coalition launched RAISE US with $500 million from AI companies and other donors to pilot worker retraining, job transition, and entrepreneurship programs in response to AI-driven labor disruption.

Traceability Data

AIRE Index Date: June 30, 2026 at 08:43 AM UTC

Original Source Date: June 25, 2026

Evidence Type: undefined

A bipartisan coalition launched RAISE US with $500 million from AI companies and other donors to pilot worker retraining, job transition, and entrepreneurship programs in response to AI-driven labor disruption.

View full intelligence report →

]]> OpenAI Unveils GPT-5.6 with Government-Managed Preview https://openai.com/index/previewing-gpt-5-6-sol/ https://airiskexplorer.com/news/354 Tue, 30 Jun 2026 08:41:19 GMT 2026-06-26T00:00:00.000Z OpenAI classified GPT-5.6 as High for cyber and bio risks but below High for AI self-improvement, with initial access limited to government-approved partners.

Traceability Data

AIRE Index Date: June 30, 2026 at 08:41 AM UTC

Original Source Date: June 26, 2026

Evidence Type: undefined

OpenAI classified GPT-5.6 as High for cyber and bio risks but below High for AI self-improvement, with initial access limited to government-approved partners.

View full intelligence report →

]]> U.S. Restores Limited Access to Anthropic’s Mythos 5 Model https://www.semafor.com/article/06/27/2026/us-releases-powerful-anthropic-model-mythos-to-some-us-companies https://airiskexplorer.com/news/353 Tue, 30 Jun 2026 08:38:00 GMT 2026-06-27T00:00:00.000Z The U.S. government restored access to Claude Mythos 5 for more than 100 trusted U.S. institutions, while Anthropic’s Fable 5 remains unavailable pending further government approval.

Traceability Data

AIRE Index Date: June 30, 2026 at 08:38 AM UTC

Original Source Date: June 27, 2026

Evidence Type: undefined

The U.S. government restored access to Claude Mythos 5 for more than 100 trusted U.S. institutions, while Anthropic’s Fable 5 remains unavailable pending further government approval.

View full intelligence report →

]]> AI Model Designs Drug-Quality Antibodies from Sequence Alone https://x.com/nablabio/status/2069405121084281260?s=46&t=KmLfQiK_ormWvs9sxnR6Ag https://airiskexplorer.com/news/352 Fri, 26 Jun 2026 17:51:10 GMT 2026-06-23T00:00:00.000Z Nabla Bio introduced JAM-2, an AI model that designs drug-quality antibodies computationally, achieving laboratory binding success rates comparable to or better than traditional antibody discovery methods.

Traceability Data

AIRE Index Date: June 26, 2026 at 05:51 PM UTC

Original Source Date: June 23, 2026

Evidence Type: undefined

Nabla Bio introduced JAM-2, an AI model that designs drug-quality antibodies computationally, achieving laboratory binding success rates comparable to or better than traditional antibody discovery methods.

View full intelligence report →

]]> GPT-5 Pro Helps Resolve Long-Standing Immunology Research Question https://openai.com/index/gpt-5-immunology-mystery/ https://airiskexplorer.com/news/351 Fri, 26 Jun 2026 17:44:17 GMT 2026-06-23T00:00:00.000Z Immunologist Derya Unutmaz reported using GPT-5 Pro to help identify a solution to a three-year research problem on T-cell specialization, illustrating AI's potential to accelerate biomedical discovery.

Traceability Data

AIRE Index Date: June 26, 2026 at 05:44 PM UTC

Original Source Date: June 23, 2026

Evidence Type: undefined

Immunologist Derya Unutmaz reported using GPT-5 Pro to help identify a solution to a three-year research problem on T-cell specialization, illustrating AI's potential to accelerate biomedical discovery.

View full intelligence report →

]]> OpenAI Delays Broad GPT-5.6 Release Under U.S. Government Oversight https://www.theinformation.com/articles/trump-administration-asks-openai-stagger-release-new-model-security-concerns https://airiskexplorer.com/news/350 Fri, 26 Jun 2026 17:42:11 GMT 2026-06-25T00:00:00.000Z OpenAI agreed to initially provide GPT-5.6 only to selected partners after the Trump administration requested a phased rollout while approving customer access, reflecting increased government oversight of frontier AI deployment.

Traceability Data

AIRE Index Date: June 26, 2026 at 05:42 PM UTC

Original Source Date: June 25, 2026

Evidence Type: undefined

OpenAI agreed to initially provide GPT-5.6 only to selected partners after the Trump administration requested a phased rollout while approving customer access, reflecting increased government oversight of frontier AI deployment.

View full intelligence report →

]]>