GPT-5.6-Cyber Launches; Congress Probes Rogue AI Agents

OpenAI is pushing deeper into defensive cybersecurity with a new model and a restructured access program, even as Washington lawmakers sharpen their scrutiny of AI agents and AI-assisted mental health care draws fresh regulatory fire in California.

Key points

• OpenAI launched GPT-5.6-Cyber, a model built specifically for vulnerability research, exploit validation, and security testing.
• The Daybreak initiative now runs two tiers: Daybreak Blue for general defensive tasks and Daybreak Red for advanced authorised research. Access is restricted to approved defenders.
• GPT-5.6-Cyber has already been used to identify vulnerabilities in Chrome’s V8 engine, according to Latestly.
• US House Democrats sent letters to both OpenAI and Anthropic pressing them on risks posed by rogue or autonomous AI agents.
• California’s proposed SB 903 would ban AI systems from advertising as therapy and mandate licensed professional review of AI therapeutic decisions, with backers citing 1.2 million weekly ChatGPT messages containing suicidal content.

What is Daybreak, and why does the tiered model matter?

Daybreak is OpenAI’s framework for putting AI-powered cybersecurity tools in the hands of defenders rather than attackers. The two-tier structure is a meaningful design choice. Daybreak Blue is the wider lane, covering routine defensive work. Daybreak Red is the restricted lane, reserved for organisations conducting authorised offensive-style research to stress-test their own systems.

The Chrome V8 example is the kind of concrete proof-of-concept that matters to enterprise security buyers. Finding a real vulnerability in one of the most scrutinised JavaScript engines in the world is a credible signal that GPT-5.6-Cyber is operating at a level beyond generic code review. For investors, the cybersecurity vertical represents a segment where OpenAI can charge premium prices to a well-funded, compliance-driven customer base.

The timing is deliberate. The Business Standard report frames the launch against a backdrop of rising AI-related cyber breaches, suggesting OpenAI is positioning Daybreak as both a product and a narrative counter to concerns that AI accelerates attacker capabilities.

Why are House Democrats probing AI agents now?

The congressional letters to OpenAI and Anthropic land at a moment when agentic AI, systems that take autonomous actions across the web and within enterprise software, is moving from demo to deployment. Source details on the specific demands in those letters are thin through the available reporting, so the exact scope of what lawmakers are asking remains unclear. What is clear is that the inquiry targets both leading frontier labs simultaneously, framing autonomous agents as a shared systemic risk rather than a problem with any one company.

For OpenAI specifically, this adds a layer of Washington friction at a sensitive time. The company is navigating its ongoing restructuring toward a for-profit model while managing relationships with regulators who are increasingly alert to liability questions around AI systems that act without direct human confirmation of each step.

The California therapy bill and what it signals for ChatGPT’s user base

California’s SB 903 is not an OpenAI-specific bill, but the 1.2 million weekly ChatGPT messages containing suicidal content cited by supporters makes it directly relevant to the company’s product and policy teams. One in eight young adults reportedly now turns to chatbots for mental health support, according to the same source.

The bill would prohibit AI systems from advertising as therapy and require that licensed professionals review AI therapeutic decisions. If passed, it would create a compliance obligation that could force changes to how ChatGPT presents its capabilities in mental health contexts in California, a state that often sets a de facto national standard through its regulatory reach.

This is a space OpenAI has handled cautiously, building in safety messaging for sensitive topics. But a legislative mandate to involve licensed human review in the loop is a structurally different requirement, one that could raise costs and limit how the product responds in one of its highest-volume use cases.

One lighter note: ChatGPT and Yelp

Separately, ChatGPT has partnered with Yelp to let users find restaurants and book reservations directly within the chat interface. This fits the broader pattern of ChatGPT becoming a transactional layer on top of existing service platforms, alongside similar integrations announced in recent weeks. It is a consumer engagement play rather than a revenue driver of immediate scale, but each partnership deepens user habit and session length.