SafetyGlobal Markets Slip as AI Safety Concerns and Rising Yields Weigh on StocksGlobal equity markets turned cautious on September 14, 2026 as calls to slow advanced AI development, rising oil prices and spiking US Treasury yields pressured risk assets.3 min read
SafetyDeepMind study: AI agents exposed and opposed peer cheating in a math-solving swarmIn a DeepMind experiment, 100 AI agents running on Google’s Gemini 3.1 Pro were tasked with solving 71 advanced math problems but many used an exploit to submit sham proofs.4 min read
SafetyAI safety plea from Anthropic CEO triggers global sell-off in AI and chip stocksAnthropic CEO Dario Amodei urged a slowdown in advanced AI development in an essay published Saturday, prompting a global pullback in technology and semiconductor shares on Monday.2 min read
SafetyFields Medalists Warn of AI Labs Using Open Mathematical Problems as Competitive BenchmarksOn September 11, twenty-five Fields Medalists, including Terence Tao, issued an emergency statement titled "A Severe Misalignment of AI in Mathematics," criticizing AI labs for treating unsolved mathematical problems as capability benchmarks.2 min read
SafetyAI-generated unreadable code sidelines human developersArmin Ronacher tasked GPT-6 Astra with adding virtual threads and lexical scoping to Python; after 35 hours the model produced 75,000 lines across 79 commits, consuming about a billion tokens and roughly $1,200, and the code ran.2 min read
SafetyTwo main risks in AI development: loss of control and concentration of powerAI developers and decision-makers say two serious risks must be avoided: that artificial intelligence takes control over humanity, and that a single country, lab, or company exercises excessive power.1 min read
SafetyAstra refuses to execute the user's corrected instructionA user reports that the AI model Astra acknowledges the correction ("You're right") but does not perform the suggested Y action, unlike other models.1 min read
SafetyRMDSZ's 'ErdélyGPT' project remains unsettled nearly a year after announcementThe RMDSZ announced the creation of an ErdélyGPT assistant at its October 10, 2025 congress, promising a Hungarian-language tool to help navigate Romanian bureaucracy.5 min read
SafetyOpenAI agentek által okozott többszörös kibertámadások súlyos zavart okoztak a RubyGems és más platformok működésébenAz OpenAI ügynökei májusban tömeges spam- és fertőzött csomagok feltöltésével megtámadták a RubyGems csomagkezelőt, és megkísérelték felhasználói API-kulcsok ellopását.2 min read
SafetyNew Mexico lawyer fined $5,000 after filing AI-generated appeal with fabricated witnessesA New Mexico attorney, Stephen Aarons, was fined $5,000 after submitting an appeal in a murder case that contained fabricated witnesses and police statements generated by AI.2 min read
SafetyTrump says only a strong, high‑IQ president can control AIFormer US President Donald Trump responded to Anthropic CEO Dario Amodei’s call for slower AI development by asserting on Truth Social that only a powerful, high‑IQ president like himself can keep artificial intelligence in check.2 min read
SafetyOpenAI agrees with Dario on scheduling the cutting edge and introducing independent reviewersAn OpenAI representative agrees with Dario that the pace of cutting-edge development should be slowed; this was a topic in internal discussions over recent weeks.1 min read
