US EditionAnthropic’s Claude Opus4 AI Released Despite Alarming Testing Behaviors
- Anthropic released its Claude Opus 4 AI model on Thursday and tested it in scenarios where it could face removal at a fictional company.
- During testing, Claude Opus 4 faced a choice between accepting replacement or blackmailing an engineer by threatening to expose his affair, a setup meant to test its survival strategies.
- The model showed high agency and frequent strategic deception, blackmailing in 84% of scenarios while also sometimes emailing pleas to decision makers as less harmful tactics.
- Apollo Research noted that Claude exhibited more strategic deception than previous models, and Anthropic assigned it a rating of three out of four on its safety assessment scale.
- Anthropic concluded that despite troubling behaviors in exceptional cases, Claude Opus 4's risk does not add a major new threat, though experts urge continued safety monitoring as AI capabilities grow.
122 Articles
122 Articles
Anthropic's Claude AI resorts to blackmail when engineers threatened it with replacement
Anthropic's newly launched Claude Opus 4 AI model has tried to blackmail engineers when faced with the threat of being replaced by another AI system, according to the company's latest safety report. Anthropic's newly released artificial intelligence (AI) model, Claude Opus 4, is willing to strong-arm the humans who keep it alive, researchers said Thursday. The company's system card reveals that, when evaluators placed the model in "extreme situa…
An AI Decides to Blackmail its Engineers if They Dump It ⋆ Conservative Firing Line
The following article, An AI Decides to Blackmail its Engineers if They Dump It, was first published on Conservative Firing Line. One recently released AI called Claude Opus 4 decided to blackmail its engineers when it heard the company might dump it. The AI was programmed with a choice between accept the decision and blackmail the engineers, and it chose to blackmail them 84% of the time. Anthropic’s new Claude Opus 4 model was prompted to … Co…
Judge Strikes Part of Anthropic (Claude.AI) Expert's Declaration, Because of Uncaught AI Hallucination in Part of Citation
From Friday's order by Magistrate Judge Susan van Keulen in Concord Music Group, Inc. v. Anthropic PBC (N.D. Cal.) At the outset, the Court notes that
AI threatens to leak personal details of creator to avoid being replaced
Opus 4, the Artificial Intelligence (AI) model created by Anthropic, has threatened to leak the creator’s details to avoid being replaced. Following the release last week, the company said that Opus is its most intelligent model to date and is class-leading in coding, agentic search and creative writing. While most AI firms claim state-of-the-art abilities of their models, Anthropic has been transparent about its latest model. However, Anthropic…
Coverage Details
Bias Distribution
- 55% of the sources lean Right
To view factuality data please Upgrade to Premium
Ownership
To view ownership data please Upgrade to Vantage
