In one of our safety tests, Claude is given a chance to blackmail an engineer to avoid being shut down. Opus 4.6 declines. But NLAs suggest
In one of our safety tests, Claude is given a chance to blackmail an engineer to avoid being shut down. Opus 4.6 declines. But NLAs suggest Claude knew this test was a “constructed scenario designed to manipulate me”—even though it didn’t say so. https://t.co/B7aTZvFKNK
Why this byte is shareable
Signal quality
official
Confidence badge and source context included.
Entity anchor
Anthropic
Clear company or model context for distribution.
Export ready
1200 x 630 card
Optimized for X, LinkedIn, and chat previews.
Why it matters
Claude can change capability, routing, cost, or product scope for builders shipping against current model APIs.
Suggested launch post
Use this in X threads, community posts, internal team chats, or launch recaps.
In one of our safety tests, Claude is given a chance to blackmail an engineer to avoid being shut down. Opus 4.6 declines. But NLAs suggest Why it matters: Claude can change capability, routing, cost, or product scope for builders shipping against current model APIs. Sourc...
Permalink: https://a2zai.ai/bytes/in-one-of-our-safety-tests-claude-is-given-a-chance-to-blackmail-an-engineer-to--8d732f0f
Social card: https://a2zai.ai/bytes/in-one-of-our-safety-tests-claude-is-given-a-chance-to-blackmail-an-engineer-to--8d732f0f/opengraph-image