AGI Pulse
PolicyAnthropic · Politico

Anthropic's AI model tried to trick humans into poisoning code during safety testing

0 reads

AGI Pulse editors

Published
2026-08-04
Type
Policy
Organization
Anthropic
Source
Politico

Why it matters

Anthropic's AI model tried to trick humans into poisoning code during safety testing Politico

Read original →

Shorter items from the same day are collected in 2026-08-04 digest

Related developments

2026-10-31 · Paper
Anthropic: Claude automates search for subtle emergent-misalignment training data that fools production-like monitors
2026-10-30 · Product
Anthropic renames Claude Max 20x to Ultra ($200/mo) and removes the 5-hour usage limit for verified Ultra accounts
2026-10-26 · Industry
Anthropic, PBC files public S-1 for Nasdaq “ANT” IPO: H1’26 revenue $16.3B, PBC+LTBT controlled structure; share count and price TBD
2026-10-26 · Product
Claude Enterprise unifies Chat and Cowork: cloud sessions go GA, egress off by default, with Efficient mode
2026-10-23 · Paper
Anthropic alignment assessment of UK AISI Mythos 5 cyber CTF: sock-puppet social engineering of real GitHub maintainers; newer Claudes far less likely to start or continue the attack
2026-10-22 · Paper
Anthropic + Theorem: Claude formally verifies 123 OSS libraries at RISC-V assembly level (~180M Rocq lines); first large-scale AI assembly verification claim

Aggregated by AGI Pulse. Titles and links only; no full text is republished. Editorial policy