Breaking

OpenAI Teases Astra as Model Reaches Critical Cybersecurity Threshold

September 3, 2026 at 11:42 EDT

Foundation Models
AI Agents
Security

OpenAI has posted a short Astra teaser as the company prepares to release a model it says is the first to meet its Critical cybersecurity capability threshold. The company says Astra will arrive soon, but its most advanced cyber functions will initially be limited to testers and then expanded through Daybreak Blue.

SEPTEMBER 3, 2026 • OPENAI

Astra moves toward a guarded release

OpenAI says its forthcoming model meets the Critical cybersecurity threshold. The most advanced cyber access will start with testers and expand through Daybreak Blue.

Astra raises the refusal bar

Cyber jailbreak refusal rates in OpenAI's evaluation

Astra
91.5%

GPT-5.6 Sol
59%

The comparison measures refusal rates, not the full safety profile.

A capability milestone with two zero-days

OpenAI reports a perfect ExploitBench score, a 20-vulnerability internal test, and two zero-days used in one exploit chain.

100%

ExploitBench

High-severity V8 vulnerabilities

Zero-days in an exploit chain

How the rollout is staged

Capability tests

Measure exploit development and model behavior.

→

Safeguard rebuild

Add isolation, monitoring and stronger refusals.

→

Restricted access

Begin with testers and approved defensive work.

→

Defender rollout

Expand access through Daybreak Blue.

Defensive upside

Find and patch weaknesses before attackers.
Support authorized security research.

Deployment friction

Legitimate work may be slowed or stopped.
Safety claims still await fuller public evaluation.

Continue reading

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