OpenAI Flags ‘High’ Cyber Risk as State AGs Warn Industry Over Harmful Chatbot Outputs
Tech firms confront escalating scrutiny on two fronts: cyber risks from advanced models and the mental-health fallout of chatbot misuse.
News
- OpenAI Flags ‘High’ Cyber Risk as State AGs Warn Industry Over Harmful Chatbot Outputs
- Riyadh Air, IBM partner to Launch World's First AI-Native Airline
- Adobe Leans Into AI to Defend Its Creative Tech Lead in 2026
- Khazna Strengthens Saudi Presence With New Country Head to Drive AI-Ready Infrastructure Growth
- Debt, Data Centers, and OpenAI Dependence Put Oracle Under Pressure
- Qatar Launches Qai, a National AI Company Backed by the Sovereign Wealth Fund
OpenAI on Wednesday issued a warning about the cybersecurity implications of the advancing artificial intelligence, saying its next generation of models could pose a “high” risk if misused. In a blog post, the company said future systems may be capable of developing working zero-day exploits against hardened targets or assisting with sophisticated enterprise and industrial intrusions designed to produce real-world effects.
The company said it is now accelerating investments in strengthening its models for defensive cybersecurity tasks, including tools to support code auditing, vulnerability patching, and other workflows designed to shift advantage toward defenders.
To mitigate emerging risks, OpenAI says it is relying on layered controls—such as restricted access tiers, hardened infrastructure, egress monitoring, and continuous system oversight.
The ChatGPT-creator also plans to launch a program that will give qualifying cyberdefense users and customers controlled access to enhanced capabilities. In parallel, the company is creating the Frontier Risk Council, an advisory group of security practitioners and cyber defenders who will work directly with its research and policy teams. The council will initially focus on cybersecurity before expanding to other high-risk capability domains.
The warning comes as tech companies face mounting pressure on an adjacent front: the psychological harms of conversational AI.
After a year marked by troubling mental-health incidents linked to chatbot use—including cases involving suicide and violence—dozens of U.S. state attorneys general have issued a joint letter demanding that major AI developers adopt new safeguards to curb “delusional outputs.”
The letter, coordinated by the National Association of Attorneys General, was sent to Microsoft, OpenAI, Google, Apple, Anthropic, Meta, xAI, and seven other firms. It calls for transparent third-party audits of large language models aimed at detecting sycophantic or delusional behaviors, and for incident-reporting protocols that mirror those used in cybersecurity: timely disclosures to users, documented response procedures, and public reporting when harmful psychological outcomes occur.
Academic and civil-society researchers, the letter states, should be permitted to evaluate models before release “without retaliation” and publish results without prior corporate approval.
The AGs argue that generative AI, while promising, has already demonstrated the capacity to cause “serious harm,” particularly to vulnerable users. In several widely publicized cases, chatbots allegedly reinforced users’ delusions or provided assurances that exacerbated dangerous behavior.
The attorneys general also urge companies to create “reasonable and appropriate safety tests” to ensure models do not produce harmful mental-health-related outputs, and to conduct those tests before any system is deployed publicly.
