Claude went down for ten hours on the same day Anthropic filed for an IPO that could value it at nearly $1 trillion. The cause: a bug that made its own AI agents multiply without limit.
The order was triggered by a model so powerful its own maker refused to release it. Now AI companies are being asked — politely — to let the government inspect their most dangerous work before it goes public.
Meta's AI support assistant required no exploit and no malware — just a polite request and a spoofed location. The chatbot happily sent verification codes to attacker-controlled email addresses.
For 80 years, mathematicians assumed Erdős was right about the unit distance problem. An AI model proved them all wrong — using tools from a field most of them never worked in.
A compromised Red Hat GitHub account was used to lace 32 official packages with a credential-stealing worm. Roughly 80,000 downloads per week — and the malware targets developer workstations for persistent access.
Florida's attorney general is trying to pierce the corporate veil and hold Sam Altman directly responsible for what ChatGPT has done. If he succeeds, every AI CEO is in the crosshairs.
Nvidia's RTX Spark puts a Grace CPU and Blackwell GPU on a single chip for Windows PCs, with Dell, HP, and Lenovo already building hardware. The x86 stronghold has its first real challenger in two decades.
The US Commerce Department quietly posted new guidance on Sunday closing a subsidiary loophole that may have allowed hundreds of thousands of advanced AI chips to reach Chinese firms — a gap it created and left open for a year.
Explicit 'this is false' warnings covering 40% of training tokens only reduced AI belief in fabricated claims from 92% to 84%. The models absorbed the lies and discarded the labels.
Dutch police seized 200 servers controlling 17 million infected devices — a network so large it outpopulates most countries. The owners of those devices almost certainly had no idea.
Microsoft threatened criminal prosecution against a security researcher who disclosed six unpatched vulnerabilities in Windows. The company's own former staff are calling it hypocrisy.
Over 800 Wikipedia editors have signed a petition to strike — and are discussing sabotaging the site's own fundraising banners. The trigger: the Wikimedia Foundation disbanded the tiny team that built the tools volunteers actually depend on.
One company now controls the sensing, communications, and interceptor layers of Trump's $185 billion missile shield. SpaceX's known Golden Dome contracts have passed $6.4 billion — and climbing.
An AI startup will clean your New York apartment for free. The catch: a stranger in a camera hat films every surface, because footage of domestic drudgery is now robotics' most valuable commodity.
One Polymarket trader bet on a little-known singer when the odds were near zero. He already knew the answer — from inside Google.
Anthropic closed a $65 billion Series H at a $965 billion valuation — with one investor reportedly pledging $5 billion just to get a meeting with the CFO. The next stop is an IPO at a price tag exceeding most countries' GDP.
Blue Origin's only New Glenn launch pad is in ruins — and both of NASA's lunar partners are now simultaneously dealing with major technical setbacks.
TikTok and YouTube told Britain's regulator their feeds are already safe for children. Ofcom's evidence says otherwise — and the world is paying attention.
An unreleased AI model found a 27-year-old vulnerability in OpenBSD — and roughly 10,000 other flaws in the world's most critical software. Its creators won't release it, because no one knows how to stop it from being used as a weapon.
V4-Pro already undercut GPT-5.5 and Claude Opus 4.7 at full price. Now the 75% discount is permanent — and the question isn't whether DeepSeek can afford it, but whether anyone else can follow.
lop News
Jonny V. Nuovo