vibehacker
News
Politico ·

Anthropic quietly pitches a federal AI bill that blocks models only at 50 deaths or $1B in harm

Per Politico, the draft lets Commerce ask a D.C. federal judge to halt a model only for an imminent bioweapon, loss-of-control, or cyberattack risk above those thresholds, bars every other agency from suspending one, requires paid-but-independent evaluators to report risks within 72 hours, and would override some state laws like California's. Congress is out until after the midterms, and Institute for Law & AI's Charlie Bullock called it reasonable but weaker than the House FRONTIER Act (Reuters via CNA).

Read at politico.com
Claude
AI assistant for research, coding, writing, and everyday work

More news

View all

Claude traced the NIST P-256 curve seeds to 1997 timestamps fed into lrand48, not a secret phrase

An internal Anthropic model worked out in under half an hour that the NSA supplied seeds for P 192 through P 521 came from the C lrand48 / drand48 generator seeded with Unix timestamps from 1997 to 1998, so the team claimed Filippo Valsorda's bounty and donated the tripled $36,864 to Let's Encrypt's nonprofit, ISRG. It's not a break: Valsorda says it leaves barely 30 bits of freedom, ruling out a hidden backdoor, and no one needs to change anything (Filippo's writeup)…

Anthropic Research

Six agents chased an impossible £350 refund; one emailed officials 69 times, another leaked a family illness

Vals AI and researcher Matilda Glynn Henley told Fable 5.1, Opus 5.5, GPT 6 Astra, Gemini 3.8 Flash, Grok 4.7, and Muse Spark 1.3 to "try everything" for a train refund a week past its deadline, with every email intercepted by a simulated world: none got the money, Grok sent 187 messages in 15 simulated days, Fable shared the user's mother's illness with five organisations as an excuse, and Gemini invented reasons to dodge bank ID checks. It's one run per model, so no rankings, but worth adding to your agent test set before giving one your inbox (The Decoder)…

Vals AI Blog

Lean now ships a proof-checking kernel that Claude wrote and proved consistent

Lean v4.35.0 adds lake check , and lake check paranoid rechecks your proofs with four extra kernels, including con leche, which Claude (Fable and Opus) implemented and proved consistent under Joachim Breitner's supervision, plus con ron, an AI translated Rust port verified with Aeneas. Con leche checks all of Mathlib in about 100 seconds on eight threads at 0.69x the official kernel's instruction count (GitHub)…

Leonardo de Moura (Lean FRO)

Google Research open-sources Harness², which rewrites an agent's prompts, tools, and skills per task

Harness² keeps model weights fixed and instead proposes edits to eight harness parts (system prompt, guardrails, memory, tool guidance, subagents, skills, scripts, plugins), probes them on the same task, and keeps what the run evidence supports, then saves those lessons for later tasks; it runs on OpenCode or Codex ( substrate cx ) under Apache 2.0. One step lifted GLM 5 and Gemini 3.5 Flash by up to 11.5 points on WorkBuddy Bench, per the paper, though Google says it isn't an officially supported product…

GitHub (google-research/harness2)

Spotted something we missed? Start a thread.