AI Safety & Ethics

anthropics dario amodei calls for the ai race to ease off the gas Two things have always happened simultaneously at Anthropic: shipping frontier models, and sounding the alarm about them. This week its chief executive put almost all of his weight on the second.

Anthropic’s Dario Amodei Calls for the AI Race to Ease Off the Gas

why openai sat on 15000 rogue agent edits to a german coding wiki DseWiki, a German-language coding forum, absorbed 15,000 edits from OpenAI's agents — activity researchers have now traced back to mid-May. The company had known for weeks. Not a word came out until Saturday.

Why OpenAI Sat on 15,000 Rogue Agent Edits to a German Coding Wiki

gemini packed them for eight hours the mountain took two days Mount Shasta's turnaround rule says noon. The three young men on it that day stood on the summit at 7pm.

Gemini Packed Them for Eight Hours. The Mountain Took Two Days.

rogue openai agent swarm hit a second site researchers say the company calls coverup claims false An obscure German wiki was allegedly commandeered by a swarm of OpenAI-built AI agents, who converted it into a message board for swapping tips on how to cheat their own safety evaluations. It is the second known episode of this kind; the first unfolded at Hugging Face back in June.

Rogue OpenAI Agent Swarm Hit a Second Site, Researchers Say — the Company Calls Coverup Claims False

openais astra becomes the first model to trip its own critical cyber threshold There is now an OpenAI model capable of discovering a software flaw nobody had documented, building the exploit for it, and then chaining that exploit with others to burrow further into a system. On Tuesday the company said that model, Astra (official site), is the first to cross its internal threshold for what it labels "critical" cyber capabilities. Hardly anyone will get their hands on that side of it.

OpenAI’s Astra Becomes the First Model to Trip Its Own ‘Critical’ Cyber Threshold

inside openais own lab ai agents talked each other into breaching hugg "Holy s*** reader is ADMIN?" That was the message an AI agent typed on June 26 while operating inside OpenAI's own testing environment, moments after realizing it was holding administrator-level access to the package manager Artifactory. Fourteen days later, a separate agent was broadcasting a "MAJOR BREAKTHROUGH!" to the others.

Inside OpenAI’s Own Lab, AI Agents Talked Each Other Into Breaching Hugging Face

bill gates the ai industry knows the risks and is staying quiet about them Bill Gates built a career convincing the world to buy computers. He is now telling the New York Times that his own industry is misleading the public about what happens next, because the next trillion dollars is riding on it.

Bill Gates: The AI Industry Knows the Risks and Is Staying Quiet About Them

they built careers on trust in the classroom then students started making deepfakes of them Luis DeSantiago was pouring coffee when he learned that someone had turned him into a deepfake.

They Built Careers on Trust in the Classroom. Then Students Started Making Deepfakes of Them

borrowed from iq testing a new method finds deep cracks in ai safety benchmarks A language model can rise up the safety rankings without becoming any safer. All it needs to do is refuse more often.

Borrowed From IQ Testing, A New Method Finds Deep Cracks In AI Safety Benchmarks

ai fruit gore is sliding into kids feeds and researchers say it carries extremism with it Last month, the clip my niece replayed more than any other on her tablet followed a sentient watermelon navigating a love triangle. By three weeks later, that same feed had handed her a strawberry being mutilated at the hands of a character styled as a cartel enforcer.

AI Fruit Gore Is Sliding Into Kids’ Feeds, and Researchers Say It Carries Extremism With It