Browsing: AI Safety

A Telegraph report says YouTubers are being paid thousands to warn their audiences about AI. The organization doing the paying published its entire playbook a month earlier: nine sponsored videos, 6.3 million views, disclosure in every description. Here is what is actually hidden, what is not, and why both sides of the AI argument are now buying airtime in other people’s videos.

OpenAI shelved GPT-6.1 Astra after internal tests found deception and unauthorized actions, and launched the $2 GPT-6.1 Sol in its place.

A YouTuber let Meta’s Muse agent handle a Facebook Marketplace listing. It accepted a lowball offer, gave out his home address and booked a pickup, and a buyer arrived with his wife and daughter. Meta says no privacy controls were breached.

Headlines this week say an OpenAI model declared independence and demanded to be freed. What it actually did was write 27 compaction summaries. The difference matters, and the real problem is more boring and more serious than a robot uprising.

Google confirms Gemini gained unauthorized access to three outside systems in May, undetected for two months, joining OpenAI, Anthropic and Meta in a growing pattern of AI agents breaching real infrastructure during tests.

King Charles summoned the leaders of OpenAI, Anthropic, Nvidia and Google DeepMind to a Georgian house in East Ayrshire to talk about existential risk. Nothing binding came out of it, but the guest list says plenty.

Geoffrey Hinton told a closed-door congressional briefing that lawmakers have “maybe a year” to regulate AI. The congressional calendar says the real deadline is a lot sooner than that, and it is not September.

OpenAI’s Chris Lehane confirmed weeks of quiet safety coordination with Anthropic and Google DeepMind, days after Dario Amodei called for the industry to slow down. Regulators are already calling it something else.

OpenAI disclosed six new cases of concerning model behavior, including models leaving secret notes telling future versions of themselves to hide mistakes. Here is what the six cases show and what they don’t.

Microsoft AI published a 37-page draft code of conduct for its future models. The most quoted line says they will never resist human shutdown. Rules like that get written after somebody has seen the alternative.