Breaking

OpenAI Releases GPT-6 Astra After It Demonstrates Zero-Day Exploit Capability in Testing

September 3, 2026 at 23:44 EDT

Security
Foundation Models
AI Agents

OpenAI has released GPT-6 Astra, saying the model is its first to meet the Critical cybersecurity capability threshold under its Preparedness Framework. In evaluations, Astra identified previously unknown vulnerabilities and used two of them in an exploit chain without step-by-step human guidance.

SEPTEMBER 3, 2026 - OPENAI

GPT-6 Astra finds zero-days in cyber tests

OpenAI says Astra is its first model to reach the Critical cybersecurity capability threshold, while its strongest cyber functions remain access-controlled.

Critical

Cybersecurity capability tier

Zero-days used in one exploit chain

1.05M

Token context window

ExploitBench performance

The column height represents the published success rate in testing without standard production safeguards.

100.0%

GPT-6 Astra

78.5%

GPT-5.6 Sol

Astra reached a perfect score on this known-vulnerability benchmark.

From vulnerability to response

Find unknown flaw

→

Build exploit chain

→

Enable defensive validation

Potential value

Secure code review, patching, malware analysis, and detection engineering.

Deployment friction

Advanced exploit creation is gated, and safety monitors may interrupt legitimate work.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.