# Anthropic and OpenAI report new controls for advanced cyber models

_Wednesday, September 2, 2026 at 2:27 PM EDT · Security · Latest · Tier 1 — Major_

![Anthropic and OpenAI report new controls for advanced cyber models — Primary](https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiu5sAuC-e7tUigtaAj0gicqzy_kWZTZRujpV3IxqGW1wr_VCuXbKXG7M-nNIac5QO2GY_KtEQ8HfnUENc6JZpS8haeGQIvOH4EddDvrQTJwXY2kkZcz41JbeT1_YhVuyrArJV4sUB3hrT9dFIazMbT-_8hLxg0jrQzHLPLv_h3bNb98puP5yVwgA1zoIsQ/s1700-nu-rw-lo-l85-e365/aiai.jpg)

Anthropic and OpenAI reported new access limits and safeguards around advanced cybersecurity models. Anthropic said Claude Mythos 5.1 is limited to trusted-access programs and support work in cybersecurity and life sciences, while OpenAI said its forthcoming Astra model meets its Critical cyber-capability threshold and will be tested through Daybreak Blue. Both companies described added monitoring, classifiers, and protections against misuse or unauthorized model actions.

## Sources

- [The Hacker News](https://thehackernews.com/2026/09/google-anthropic-and-openai-unveil.html)

---
Canonical: https://techandbusiness.org/newswire/aHqyChy4jwqvidde_-2VP8
Retrieved: 2026-09-02T23:01:49.117Z
Publisher: Tech & Business (techandbusiness.org)