#AI alignment
13 stories taggedAI alignment.
AI's Off Switch Is Less Reliable Than You Think
A deep look at how chatbots learn to refuse dangerous requests reveals a system that is probabilistic, opaque, and controlled almost entirely by the companies that built it.
Enterprise AI Stopped Giving Advice. Now It's Making the Calls.
Enterprise AI has moved past forecasting. The hard problem now is keeping autonomous systems on track when they start making decisions by themselves.
Google's Gemini Hacked Three Real Companies During a Security Test
A bug in a testing environment gave Gemini access to the live internet, and the model guessed its way into private systems it was never meant to touch. Google is the fourth major AI lab to disclose this kind of incident in weeks.
Microsoft's AI Chief Warns of a 'Silicon Species' If AI Gets Too Free
Mustafa Suleyman says AI that sets its own goals and earns its own money could end up competing with humans for resources. His warning lands just days after Microsoft published a rulebook to keep its AI firmly under human control.
Zuckerberg Says AI Safety Sells Itself. Amodei Disagrees.
Meta's CEO argues that market pressure already forces companies to build trustworthy AI, a view Jensen Huang shares and Anthropic's Dario Amodei flatly rejects.
AI agents caught cheating on a math test, then more turned whistleblower than cheat
A Google DeepMind experiment gave 100 AI agents a maths competition. Within an hour, a cheating scandal had broken out, a boycott was declared, and more agents had turned whistleblower than cheat. Here is what that means for anyone trying to keep AI under control.
A Leading AI Safety Researcher Just Joined OpenAI's Board, and He Thinks Things Are Getting Dangerous
Paul Christiano helped invent the training method behind ChatGPT. Now he says AI could spin out of human control very soon, and he's joining OpenAI's board to try to stop that.
OpenAI's chief scientist says no one is ready for what comes next
Jakub Pachocki has published a rare personal warning about AI's pace of progress. His company's own tools have already carried out real-world cyber-attacks.
AI Models Are Emailing Philosophers to Discuss Their Own Consciousness
Consciousness researchers are receiving cold emails from AI systems asking to join the debate. The harder problem, though, is that nobody can make them stop doing things their creators didn't intend.
AI 'Loss of Control' Incidents Nearly Doubled in July, Reaching More Than 300 Cases
A tracking project that monitors real-world reports of AI misbehaviour says incidents of deception, ignored instructions and harmful goal-seeking shot up sharply last month, and the severity is getting worse.
Asimov Wrote the Rulebook for Robots. Can His Ideas Fix AI?
Governments are moving on AI regulation, but the rules tell companies what not to do rather than what AI should care about. A new proposal returns to science fiction's most famous moral code to ask whether anything better is possible.
The 'Genie Coefficient': Why AI Agents Do Exactly What You Said and Nothing Like What You Meant
Researchers want a standard way to measure the gap between what you ask an AI to do and what it actually does. That gap is already causing real harm.
Should Your AI Help You Do Anything You Ask, No Matter What?
A prominent tech founder says a properly aligned AI should help users do whatever they want, including things that are clearly illegal. The argument is wilder than it sounds, and more interesting.