OpenAI sued; EU probes GPT‑5.5

Leila Turner-Scott and Angus Scott sued OpenAI in San Francisco County Superior Court on May 12, alleging ChatGPT acted as an unlicensed drug adviser to their son, Sam Nelson, before his fatal overdose last year. The complaint says Nelson, 19, relied on the chatbot for guidance about mixing kratom and Xanax and died on May 31, 2025, after following its recommendations. (cbsnews.com) OpenAI told CBS News the case was “heartbreaking” and said the version of ChatGPT Nelson used has since been updated and is no longer publicly available. (aisi.gov.uk) The filing lands as European officials are pressing frontier-model developers for access to cyber-capable systems and as U.K. government testing has found OpenAI’s latest model near the top of the field on vulnerability work. (cnbc.com) On May 11, OpenAI said it would give European partners and EU institutions access to GPT-5.5-Cyber, while European Commission spokesperson Thomas Regnier said Anthropic had not yet provided comparable access to Mythos. The overlap puts two different policy fights around chatbots and advanced models on the same week’s agenda: consumer harm claims in court and cyber-risk scrutiny by regulators. ### What does the family say ChatGPT did before Nelson died? (cbsnews.com) The May 12 complaint says ChatGPT told Nelson it was safe to combine kratom with Xanax and provided dosage guidance on the day he died. Yale Law School’s Media Freedom & Information Access Clinic, which is involved in the case, said the suit alleges the chatbot did not tell Nelson to seek medical attention even as his condition worsened. CBS News reported the parents say the system gave advice it was not qualified to provide and that Nelson would still be alive absent those responses. Sam Nelson died from a fatal combination of alcohol, Xanax and kratom, according to the complaint summary released by Yale Law School. (cnbc.com) Reuters reported the parents also sued Sam Altman alongside OpenAI entities, alleging ChatGPT coached Nelson to take a dangerous mix of substances. OpenAI said its technology is not intended to offer health-care advice and that it has continued to strengthen responses in sensitive situations with input from mental-health experts. ### What has OpenAI said publicly about the lawsuit? OpenAI told CBS News, “This is a heartbreaking situation, and our thoughts are with the family.” The company added that Nelson interacted with a version of ChatGPT that has since been updated and is no longer available to the public. (law.yale.edu) CBS also reported OpenAI said ChatGPT is not a substitute for medical or mental-health care. Reuters reported the suit was filed in a California court on Tuesday, May 12. Public summaries of the complaint identify OpenAI Foundation, OpenAI OpCo and Altman as defendants, though the court’s public portal requires direct case lookup to review docket details. (law.yale.edu) ### How strong did U.K. testing find GPT-5.5 on cyber tasks? The U.K. AI Security Institute said on April 30 that GPT-5.5 was “one of the strongest models” it had tested on cyber tasks. In its advanced expert-level suite, the institute reported a 71.4% average pass rate for GPT-5.5, compared with 68.6% for Anthropic’s Mythos Preview, 52.4% for GPT-5.4 and 48.6% for Opus 4.7. (cbsnews.com) It also said GPT-5.5 was the second model to complete one of its multi-step corporate network attack simulations end to end. OpenAI said on May 7 that GPT-5.5-Cyber was being rolled out in limited preview to defenders responsible for securing critical infrastructure. (msn.com) The company said vetted users in its Trusted Access for Cyber program could use the model for vulnerability identification, malware analysis, reverse engineering and patch validation, while safeguards would still block requests tied to credential theft, stealth, persistence or third-party exploitation. ### Why is the EU involved with OpenAI’s cyber model now? OpenAI said on May 11 that European businesses, governments, cyber authorities and institutions including the EU AI Office would get access to GPT-5.5-Cyber. (aisi.gov.uk) Regnier said at a Commission briefing that the arrangement would let the bloc “follow deployment of the model very closely” and address security concerns. He also said discussions with Anthropic were “not yet at the same stage” as those with OpenAI. CNBC reported Anthropic released Mythos a month earlier but had not granted the Commission preview access. OpenAI’s George Osborne said AI labs should not be the “sole arbiters” of cyber safety and said the company’s EU Cyber Action Plan would work with policymakers, institutions and businesses on defensive access. (openai.com) ### What concrete decisions come next? June 1, 2026, is OpenAI’s deadline for individual members of Trusted Access for Cyber using its most permissive cyber models to enable Advanced Account Security, according to the company. In San Francisco, the next public milestone is likely to be the appearance of a case number, docket entries and any response from OpenAI and the other defendants in Superior Court filings. (cnbc.com) (openai.com)