OpenAI Safety Head Johannes Heidecke Departs as Company Reshuffles Research and Safety Teams
OpenAI's head of safety systems Johannes Heidecke told staff he is leaving the company following a reorganization that integrates safety and research.
OpenAI's head of safety systems Johannes Heidecke told staff he is leaving the company following a reorganization that integrates safety and research.
OpenAI merged 37 patches across 19 open-source projects in the first week of its Trail of Bits partnership.
Anthropic expanded Project Glasswing to 150 organizations across 15 countries, keeping its powerful vulnerability-hunting AI restricted from public.
Meta patched a critical flaw after hackers tricked its AI support chatbot into transferring email addresses and resetting passwords on high-profile.
A study by 22 public broadcasters found AI chatbots misrepresent news 45% of the time, prompting election officials worldwide to warn voters against.
Cybersecurity researchers find AI chatbots generate incorrect URLs for major brands, enabling scammers to hijack unclaimed domains and steal banking.
Google says it stopped hackers using AI to build a zero-day exploit, a milestone that confirms long-held fears about AI weaponization in cyberattacks.
OpenAI launches opt-in Trusted Contact feature that alerts designated person when ChatGPT detects potential self-harm discussions, expanding safety tools.
Mozilla reveals Anthropic's Mythos AI found 271 Firefox vulnerabilities since February with almost no false positives, signaling a major shift.
New benchmarks reveal GPT-5.5-Cyber equals Anthropic’s Mythos at finding flaws, making the dual lockdowns look less about unique risk and more about market control.
Law enforcement overwhelmed by surge of AI-created child abuse imagery, struggling to identify real victims amid synthetic content explosion.
A small group gained unauthorized access to Anthropic's powerful Mythos cybersecurity model through a private forum, raising concerns about AI security.
Mozilla patched 151 hidden bugs before Firefox 150 shipped, but leaked Anthropic Mythos model is already being used by Discord hackers to find new exploits.
OpenAI’s Trusted Access for Cyber program scales to thousands of vetted defenders, releasing GPT-5.4-Cyber as a gated defensive-only model with $10M.
Dario Amodei meets White House Chief of Staff Susie Wiles Friday, signaling thaw in patch-freeze standoff while UK bank launch proceeds.