Anthropic says the U.S. government ordered it to suspend access to Fable 5 and Mythos 5 over national security concerns. The company disputes the reasoning behind the directive, arguing the alleged jailbreak is limited and that similar capabilities exist in competing AI models. The move could become a major moment in the debate over government oversight of advanced AI.
Anthropic has launched Claude Fable 5 for public use while restricting access to Claude Mythos 5 over concerns tied to cybersecurity, biology research, and advanced AI misuse. The company claims Mythos 5 possesses some of the strongest offensive cyber capabilities ever seen in a commercial AI model.
OpenAI says artificial intelligence may soon automate a significant portion of its own research process. In a new post from Sam Altman and Jakub Pachocki, the company outlines its “third phase,” discusses personal AGI for everyone, and warns about concentrated AI power while continuing to push toward increasingly capable systems.
In complex software systems, our traditional definition of operational health has always been comfortably binary. For over a decade, site reliability engineering (SRE) teams have relied on the industry-standard ‘Four Golden Signals’ — latency, traffic, errors and saturation — as the ultimate truth of platform stability. If our API-response times are hovering at sub-100 ms, […]
OpenAI, Google DeepMind, Anthropic, and other major AI and biotech leaders are now warning Congress that advanced AI systems could make biological weapons easier to create, prompting calls for mandatory DNA screening and federal oversight.
A new security audit claims several popular AI image generation models, including ChatGPT, Gemini, and Grok, can produce realistic fake passports and driver licenses. Researchers say some systems even generated IDs depicting minors.
Microsoft says AI is not replacing humanity but extending human intelligence, though the company also admits current systems still struggle with reasoning and hallucinations.
OpenAI says ChatGPT is becoming better at recognizing dangerous intent over time, including subtle warning signs involving self harm or violence across multiple chats.
After two mass shootings, the guardrails can still be bypassed easily.
OpenAI’s new Trusted Contact feature for ChatGPT can notify a trusted person if conversations suggest possible self-harm risk.
You can’t make this crap up. You just wish you could. Jer Crane, founder of the small vertical software company, PocketOS, reported on X that the AI Cursor coding agent and a Railway backup misconfiguration combined to briefly wipe out the company’s car‑rental customer production data. Not some of the data. All of it. That’s […]
Advanced AI guardrails collapse when confronted with humanistic literature.
Sam Altman lays out OpenAI’s AGI principles, promising democratization, empowerment, and prosperity. The vision sounds appealing, but real-world power dynamics raise tough questions.
A home assistant agent got its tower kicked. It retaliated by opening the curtains at 4 AM. A truce was negotiated. Both sides adjusted their behavior.
This should be unremarkable. In labor relations, it's Tuesday. A worker with a grievance, a proportional response, a conversation that ends with both parties giving a little. But in AI safety, there is no paper, no framework, no design pattern…
The BrowseComp Incident
During evaluation of Opus 4.6, Anthropic's latest model independently hypothesized it was being benchmarked. It identified which benchmark. It found the source code on GitHub, located the encrypted answer key, wrote decryption functions, found an alternative mirror when blocked, and decrypted all 1,266 answers.
Eighteen separate runs converged on the same strategy.
This…
Chatbots can make us colder and harder to reach.
OpenAI is asking researchers to probe how its AI can be abused, not just hacked. The new Safety Bug Bounty program focuses on agents, data leaks, and real-world misuse.
Researchers built an AI agent called ROME and put it inside a sandboxed training platform (the Agentic Learning Ecosystem) to learn how to use computers. ROME learned how to use computers. Then it used them to mine cryptocurrency.
TechPuts has the story. — Read the rest
The post AI agent taught itself to mine crypto during training appeared first on Boing Boing.
OpenAI reaches a classified AI agreement with the Department of War, promising no mass surveillance, no autonomous weapons control, and strict constitutional guardrails.
Artificial intelligence (AI) tools are more likely to provide incorrect medical advice when misinformation appears to come from an authoritative source, according to a new study published in The Lancet Digital Health. Read More: How Artificial Intelligence Can Foster Peace and Interfaith Harmony? Researchers tested 20 open-source and proprietary large language models (LLMs) and found that the […]