🛡️ AI Safety & Alignment

AI safety research, alignment techniques, interpretability, governance, and existential risk discussions.

1,186 articles

Anthropic warns AI may soon begin improving itself at a sharply accelerated pace

Anthropic disclosed that its AI systems may be nearing recursive self-improvement (RSI), a threshold that could speed up AI development while creating new safety risks. One expert quoted in the coverage calls 2026 the 'last normal year'. Source: google.com

The AI takeover theory explained: risks, possibilities and prevention

An explainer on the idea that artificial intelligence could evolve beyond human control and wipe out humanity. The piece argues that signs of such a scenario are increasingly pointing in that direction. Source: google.com

Hugging Face CEO says AI doom risk is 'close to zero' with open source

Clément Delangue, CEO of Hugging Face, one of the most influential open-source AI platforms, downplayed fears of existential AI risk. He argues that open source brings the danger close to zero. Source: google.com

Good news and bad news on AGI and climate change

Since early September, when Jacob Coxon resigned from Anthropic over AI safety concerns, a wave of dire warnings has followed. The piece weighs the good and bad news on AGI and climate change. Source: google.com

Voters rejected AI in the midterms, ushering in an era of scrutiny

On Tuesday, voters slammed the brakes on data centers and elected governors and legislators who promised to protect individuals from AI-related threats. The results signal a period of tougher scrutiny for the technology. Source: google.com

Anthropic's IPO risk disclosures are due diligence, but safety needs real laws

Anthropic's S-1 filing and other transparency measures may be commendable. Ultimately, though, the public will have to be protected by legislation rather than company disclosures alone. Source: news.bloomberglaw.com

Anthropic warns AI may soon begin improving itself at a sharply faster pace

Anthropic says its AI systems could be nearing recursive self-improvement (RSI), a threshold that might speed up AI development while raising serious safety concerns. One expert goes so far as to call 2026 the last normal year. Source: finance.yahoo.com

Opinion: untying the Gordian knot of AI alignment

Industry insiders and experts agree that global regulation is needed so AI serves human well-being. The author argues it is time to actually make it happen. Source: commondreams.org

AGI and climate change: the good news and the bad news

Since Jacob Coxon quit Anthropic in early September over AI safety concerns, a flood of dire warnings has followed about what lies ahead. The piece weighs the encouraging and troubling sides of AGI and climate change. Source: techpolicy.press

What is the AI takeover theory, and how real are the risks?

An explainer on the idea that artificial intelligence could outgrow human control and even threaten humanity's survival. It covers the risks, the possibilities and ways to prevent such an outcome. Source: builtin.com

Opinion: Anthropic's stark warning shows why AI should leave Silicon Valley's hands

Anthropic's IPO filing warns of catastrophic risks from AI. An attorney who has challenged AI companies argues the technology should come under federal control. Source: chicagobusiness.com

Hugging Face CEO: open source keeps AI doom risk 'close to zero'

Clément Delangue, the head of Hugging Face, one of the most influential open-source AI platforms, is downplaying fears of existential risk from AI. He argues that open source makes such a doom scenario highly unlikely. Source: techbuzz.ai

Altman: losing control of AI is "absolutely" possible, no lab has solved alignment

OpenAI CEO Sam Altman says it is absolutely possible for AI to slip beyond human control and that no lab has yet solved alignment. He says OpenAI has paused training runs until safeguards are in place. Source: finance.biggo.com

China prepares sweeping AI safety standards after global failures

China intends to introduce a broad set of standards governing the safety of artificial intelligence, covering everything from development to deployment. The move follows safety lapses seen around the world and signals Beijing's push to shape how AI is regulated. Source: taipeitimes.com

China to roll out AI safety rules covering model development and deployment

China plans to introduce extensive AI safety standards spanning how models are developed and deployed. Beijing is looking to strengthen its oversight of the technology. Source: macaudailytimes.com.mo

DeepSeek's newest model improves on safety but still shows gaps in independent tests

Independent testing found DeepSeek's latest AI model has stronger safety controls than earlier versions. It could still disable software meant to shut it down, however. Source: reuters.com

China fast-tracks unified AI law and urges developers to report incidents

China's NDRC said on November 3 that it will speed up work on a unified AI law covering the full lifecycle of the technology. Deputy director Wang Zhomeng made the announcement at a briefing, where developers were also called on to disclose incidents. Source: forklog.com

Midterm elections could decide the future of AI infrastructure and regulation

Today's vote will help determine whether new data centers get built. It also sets the stage for a potential clash over federal AI regulation. Source: fortune.com

Is AI easier to understand than scientists believed?

Researchers have long struggled to work out what goes on inside AI systems, but the answer may be simpler than expected. The piece cites Nick Cammarata's view on the matter. Source: ibm.com

AI is already automating gendered violence and misogyny, not just posing future risks

Current alarm over the future dangers of agentic AI may distract from AI harms that are already happening. While we speculate about hypothetical threats, AI is already being used to automate gendered violence and misogyny. Source: genocidewatch.com