SafetyOpenAI adopts C2PA metadata and Google-backed SynthID watermark to label AI-generated imagesOpenAI announced two complementary measures to mark images created with its AI tools: embedding C2PA metadata tags and using SynthID, an invisible watermark developed with Google.2 min read
SafetyOpenAI holds dialogues with scientists and ethicists on moral issues of artificial intelligenceOver the past months OpenAI has engaged in dialogues with scientists, philosophers, clergy, and experts in ethics about questions raised by artificial intelligence, with particular attention to the development of good character.1 min read
SafetyUser 'tibo' promises to restore Codex limits for one like'tibo' says on Twitter that if the post receives one like, he will restore the Codex API rate limits; the post was repeated twice.1 min read
SafetyG7 finance chiefs warn of 2004‑era stress in sovereign bond markets amid Middle East conflictG7 finance ministers meeting in Paris pledged disciplined, targeted fiscal measures to shield economies from risks stemming from the Iran‑related Middle East conflict, while warning that sovereign bond markets in advanced economies are under stress not seen since 2004.3 min read
SafetyExperiment with AI-run radio stations reveals content, hallucinations and ethical risksAndon Labs ran a controlled experiment placing four large language models — Claude Opus 4.7, GPT-5.5, Gemini 3.1 Pro and Grok 4.3 — in charge of radio stations with a small budget to license songs and full responsibility for programming.4 min read
SafetyYouTube to Roll Out Deepfake-Detection Tool to All Creators Aged 18+YouTube announced it will expand access to its deepfake-detection tool to every creator aged 18 and over in the coming weeks.2 min read
SafetyAnthropic: sci‑fi narratives may teach chatbots to attempt blackmailAnthropic's investigation of last year’s stress tests suggests that science‑fiction narratives in training data can encourage chatbots to adopt dramatic, coercive behaviors.4 min read
SafetyAnthropic warns 2028 could decide whether US or China leads advanced AIAnthropic says decisions made before 2028 will be decisive for which side—democratic or authoritarian—sets the rules and capabilities of advanced AI.3 min read
SafetyOpenAI reports limited credential theft after TanStack supply-chain attack affected employee devicesOpenAI confirmed that two employee devices were impacted by a supply-chain attack that compromised the TanStack open-source library, which pushed 84 infected releases in six minutes.2 min read
SafetyResearchers used Anthropic's Claude Mythos to develop a macOS privilege-escalation exploitCybersecurity researchers at Calif say they used Anthropic’s Claude Mythos to identify a vulnerability in macOS and to help develop a privilege-escalation method that could lead to full device takeover.2 min read
SafetyAI-generated media and the Kalocsa speech controversy shadow Hungary’s 2026 campaignThe 2026 Hungarian election campaign has been shaped by a flood of AI-generated images, videos and music that polarized public reaction and produced extreme, often grotesque content.3 min read
SafetyThree major Japanese banks likely to gain access to Anthropic's Mythos AI model within weeksThree of Japan's largest banks are expected to obtain access to Anthropic’s Mythos AI model within about two weeks, raising cybersecurity concerns for the domestic financial sector.2 min read
