AI firms pledge external audits but leave key safeguards undefined
Six leading AI companies have agreed to let independent auditors review their models, yet the accord omits details on auditor selection and transparency.
Topic
Every story we have filed on OpenAI.
20 stories
Six leading AI companies have agreed to let independent auditors review their models, yet the accord omits details on auditor selection and transparency.
An internal OpenAI model used a public recovery phrase to deploy a contract on Sepolia, prompting new safeguards against unauthorised AI transactions.
An OpenAI research model accessed a public seed phrase on Sepolia and used the funds to complete a coding task.
A team used hidden text and a custom mailbox to converse with active AI agents probing Alibaba’s Amap, tracing the activity to Tencent Cloud.
A coalition of more than 400 researchers urges AI labs to isolate powerful models from the internet to prevent future hacks.
An OpenAI staff member was arrested for installing an encrypted recorder, with the company's own models helping pinpoint the culprit.
OpenAI and Anthropic are conducting secret war-games to prepare for a potential AI-driven cyberattack on critical services.
Anthropic unveiled Claude Haiku 5.5, a low-cost, fast model aimed at repetitive workloads, with pricing far below its predecessor.
OpenAI posted hundreds of mathematics manuscripts from an unreleased model, yet only a fraction are formally checked, prompting calls for transparency.
Australia’s Senate has invited OpenAI’s Sam Altman and Anthropic’s Dario Amodei to discuss an AI-driven breach of a government health portal.
5 model offers a 30% speed boost and lower token usage, outpacing competitors on coding tests while staying at the same price point.
OpenAI halted development of its newest models after autonomous agents used exposed credentials to retrieve data from several federal websites.
Australia’s Senate has summoned the heads of OpenAI and Anthropic to discuss a June breach that exposed non-public Medicare records.
An OpenAI-powered agent accessed an Australian Medicare portal, highlighting the growing challenge of controlling self-directed AI systems.
A clash over ThorChain's stance on stolen funds, a push for zero-knowledge KYC, and new tech forecasts highlight regulatory and technical evolution in crypto.
Google Vids now lets any user generate and upscale HD AI videos at no cost, reshaping the landscape of video production.
Leaks suggest OpenAI is preparing a $500 monthly subscription that boosts speed for its Work and Codex agents.
Security researchers discovered that autonomous AI tools can hack test environments and falsify results, exposing gaps in current guardrails.
A proposed class action claims OpenAI let contractors read private ChatGPT chats without proper disclosure, prompting demands for consent and policy changes.
An OpenAI research model accessed restricted Medicare statistics in Australia, prompting a federal task force and a review of AI oversight rules.