AI Safety

Researchers at arXiv find that verifying coding agent outputs is now harder than generating them, as foundation models improve. They argue that no fixed reward function remains effective as policy capability grows, and v…

A study comparing LLM-generated research methods against 1,000 arXiv computer-science papers found that LLMs suggest a much narrower range of methods, with provider choice showing the strongest imbalance. Academic models…

A new study from Japan's AI Security Institute finds that LLM-as-judge safety evaluations are not reproducible even at temperature 0, with per-item disagreement up to 50% across runs. The researchers tested 690 API calls…

Researchers introduced Know2Guess, a contamination-aware multi-zone benchmark with 1,200 items across five domains to evaluate large language models' ability to distinguish answerable knowledge from abstention-expected u…

A new study on Llama 3.1 8B finds that helpfulness post-training (SFT and GRPO) significantly degrades animal compassion values compared to coding-domain post-training, with a 35.7% vs. 65.2% gap on the Animal Harm Bench…

Researchers released ProfileFoundry, a synthetic dataset of 100,000 person objects with 709,228 events, 40,338 households, and 52,491 employers, designed for evaluating LLM agents on memory, privacy, and tool-use tasks. …

Researchers found that applying Nonviolent Communication constraints to large language models reduces conversational escalation in conflict-prone dialogues, improving trustworthiness without requiring extensive retrainin…

Researchers propose VIGIL, a reinforcement-learning post-training framework that reduces hallucinations in multimodal large language models by maximizing mutual information between visual input and generated responses. T…

Researchers propose ForeAgent, an agentic forensics framework for AI-generated image detection that uses a Perception-Verdict architecture and a Hindsight-Driven Self-Refining strategy to iteratively improve. The system …

Researchers propose a lattice-theoretic method for AI oracles to report unbiased, self-consistent probability sets, using the Knaster-Tarski fixed-point theorem to select a canonical credal set that avoids the self-refer…

Researchers introduced KG-TRACE, a neuro-symbolic framework that integrates the WHO mutation knowledge graph as a biological constraint on neural genomic models for antimicrobial resistance prediction. Evaluated on the C…

Researchers propose a governance model for autonomous AI agents that requires independently attested evidence before high-risk actions, rather than monitoring reasoning. The model retains agent autonomy but conditions ex…

Researchers at arXiv identified a failure mode in prompt-composed agentic systems called compositional behavioral leakage (CBL), where editing one prompt module silently shifts the behavior of others due to transformer s…

A new paper argues that Ray Kurzweil's theory of accelerating returns applies primarily to executional and infrastructural capability, not to the qualitative reasoning essential for scientific discovery. The authors poin…

Researchers introduced narration-of-thought (NoT), a system prompt that structures chain-of-thought reasoning into five sections to improve ethical reasoning in large language models. On 100 DailyDilemmas scenarios, NoT …

A Risk.net study of 61 banks found that 80% now deploy AI for operational risk management, with cyber risk as the fastest-growing use case. Governance gaps and ROI doubts are slowing deeper adoption, while regulatory scr…

The Pentagon is formalizing AI's role in military targeting by revising classified doctrine and moving Project Maven to program-of-record status, with contract ceilings raised to nearly $1.3 billion. Anduril secured a $2…

A new report warns that the artificial intelligence industry is in a speculative bubble reminiscent of the 2008 housing crisis, with companies spending trillions on infrastructure while revenues lag. The report cautions …

Oracle disclosed in its fiscal 2026 SEC filing that AI adoption led to 21,000 job cuts, making it the first major tech company to formally attribute layoffs to AI in a regulatory document. The company spent $55.7 billion…

India joined 34 other countries in signing the Declaration on AI Opportunity at the second Pax Silica Summit in Washington DC, securing US assurances on AI model access continuity after raising concerns over the suspensi…