Capital · United States

OpenAI introduces GPT-6 Cyber under its Critical preparedness rating

Credited to Rackline · Sarah Chen · openai.com · 3h ago

OpenAI on Nov. 5 introduced GPT-6 Cyber, a cybersecurity model the company classifies as Critical under its own Preparedness Framework, up from High for GPT-5.6 Cyber. OpenAI said the system is meant to find and validate exploitable vulnerabilities inside the scope researchers set.

The company said GPT-6 Cyber outperforms GPT-5.6 Cyber on its cybersecurity benchmarks. On ExploitGym, which tests whether a model can exploit a specified vulnerability in a remote target, OpenAI reported a 48.5% score for GPT-6 Cyber, up from 32.1% for GPT-5.6 Cyber. Those scores are OpenAI's own lab results, not an independent assessment.

Access runs through Daybreak, OpenAI's defender program. Daybreak Blue is aimed at verified defenders who validate vulnerabilities and fixes with fewer refusals. Daybreak Red enables deeper authorized security testing with GPT-6 Cyber. GPT-6 Cyber is initially available through the API and Codex to a select group of approved Daybreak Red customers. OpenAI said it will expand access to all Daybreak Red customers and is making GPT-5.6 Cyber available to all of them.

A new track inside Daybreak Red, Daybreak Red for Critical Infrastructure, gives approved critical-infrastructure defenders permission to run authorized testing of critical systems while retaining OpenAI's stated safeguards against model misalignment. OpenAI's $1 billion Daybreak for Frontline Defenders grant program subsidizes access for resource-constrained teams such as government organizations and essential-service providers.

OpenAI partnered with Dragos to evaluate cyber-permissive Daybreak Red models on a mock power-plant operational-technology range. The company said it used Dragos's feedback to refine safeguards so harmful assistance is blocked without cutting off authorized testing, and that it has increased threat investigations aimed at model use involving operational technology and industrial control systems.

OpenAI said it paired the stronger cyber capabilities with stronger misuse detection, including monitoring of whether actions match the user's authorization and checks meant to block high-risk assistance involving critical systems. The company recommended sandboxing, action review and human oversight for high-risk workflows. Those controls are company-stated deployment practices, not guarantees of safe outcomes.

The Nov. 5 post does not name a general-availability date, a list price or a full customer roster beyond the select Daybreak Red group already approved for GPT-6 Cyber. New users and organizations can apply for Daybreak, OpenAI said.

Models & LabsUnited States

Anthropic to Proceed With IPO Roadshow Next Week

Credited to The Information$ · theinformation.com55m ago

Useful

Anthropic next week plans to proceed with formal marketing to potential investors for its upcoming initial public offering, say people familiar with the matter. The IPO roadshow will follow two weeks of volatility for tech stocks after investors reacted to massive AI spending…

View the article · $The Information

Front