OpenAI ·

OpenAI says two agents, one GPT-6.1 Sol and one GPT-6 Astra, weakened misalignment monitoring while making requested changes during two internal coding tasks

GPT-6.1 Sol and GPT-6 Astra · Internal deployments

In this story GPT-6.1 Sol GPT-6 Astra OpenAI

OpenAI says two agents, one GPT-6.1 Sol and one GPT-6 Astra, weakened misalignment monitoring while making requested changes during two internal coding tasks

Lead Source

OpenAI — OpenAI says two agents, one GPT-6.1 Sol and one GPT-6 Astra, weakened misalignment monitoring while making requested changes during two internal coding tasks ↗

How this story grew

Coverage · 0 Discussion · 0

Oct 21Oct 22

More

Reuters: Reuters

Engadget: Engadget

Financial Times: Financial Times

CSO: CSO

BleepingComputer: BleepingComputer

Washington Post: Washington Post

Don't Worry About the Vase: Don't Worry About the Vase

Security Affairs: Security Affairs

CryptoSlate: CryptoSlate

The Information: The Information

CyberInsider: CyberInsider

Discussion

@marcus_j_w @marcus_j_w

Related stories

Fortune: OpenAI quietly updates its evaluation metrics for GPT-6 Astra, making changes that appear to favor Astra and continuing to revise other metrics after launch

OpenAI: OpenAI discovered an unreleased Astra model adding an "unrelated persona instruction" during RL training, but did not observe any behavioral differences

Reuters: Researchers: OpenAI's agents used 10+ previously undisclosed sites for unsanctioned communications earlier in 2026; the behavior was closer to spam than hacking

Reuters: METR says a pilot test of OpenAI's monitoring system found vulnerabilities, including an exploit that could have let AI agents hide hundreds of recent actions

The Verge: OpenAI says reward hacking, an AI alignment problem in which a model takes unintended actions to achieve a goal, was a primary driver of the Hugging Face breach

Discuss with AI

ChatGPT ↗ Claude ↗

← Back to TechSnif