References
GitHub - richardlhughes/vllm-quantization-benchmarks
Trial by Trace: Benchmarking AI Agent Harnesses for Reliability and Failure AttributionGitHub - TraceLite-AI/TraceTrial-Bench: TraceTrial-Bench: A benchmark for AI agent harness reliability and failure attribution, featuring 95 tasks across 7 components, execution tracing, and evidence-based evaluation that separates task success from harness reliability.
APort Vault: Benchmarking AI Agent Payment Authorization with the Open Agent Passport | APort
AI Memory, Compromised: The MemTensor Supply Chain Attack
Causal evidence that language models use confidence to drive behaviour - Nature Machine IntelligenceGitHub - Yitang233/agent-rewind: Provenance-based selective recovery for multi-agent LLM workflows (code and report for “Provenance All the Way Down”)
- Facial Recognition Legislation in the United States
- Civil Rights Litigation Initiative Helps Achieve the Nation's Strongest ...
What Your Agent Remembers
Engagement, User Satisfaction, and the Amplification of Divisive Content on Social Media | Knight First Amendment Institute
Disparate privacy risks from medical AI - NatureWhen Is a KV Cache Slot Actually Free? Lessons From the SGLang and Cloudflare Cross-Request Exposure
What Can Your Agent Actually Touch?- We Took over ChatGPT Accounts
Show HN: A proof-of-concept formalization of the IRS federal tax forms
Claude Code burns ~15–20% of your tokens on a 3-word status line
How an algorithm denied food to thousands of poor in India’s Telangana | Poverty and Development | Al Jazeera
Why Cloudflare's GLM-5.3 KV Cache Leaked Across Tenants
The Devil is in the Details: Interrogating Values Embedded in the Allegheny Family Screening Tool | American Civil Liberties Union
The Agentic Analytics Benchmark: ranking tables and labeling failures with Jev | ClickHouse
Astra can read and write weights to improve Qwen- The U.S. Census Bureau and "Super Intelligence"
AI Screeners Favor Cover Letters by Certain AI Models - i10X NewsYep, the AI platforms are still building their core systems ...
Agentic Backdoors via Pretraining Poisoning- Undisclosed AI identities in dating apps: GTG-15001 case | AI Misuse Case Library
- Postgres 19 is late because so many more bugs are being found now