Are we in a horror movie?

AI Safety

There is a movie sub-genre, within the horror genre, where a group of people end up in a cabin in the woods. It has become so iconic that we rarely reject the premise and just accept that a group would willingly gather in such a place, with a murderous monster no doubt looming in the forest nearby.

The thing about these movies is that as they unfold they are full of moments in which a reasonable person would simply back away, turn around, go home, get the hell out of there. The clues are everywhere. But they don’t. Otherwise there is no movie.

In our current version of this movie, in which AI is about to run amok and possibly kill us all while trashing our democracy, our health, our economy, the clues are also all around us. And you have to ask yourself, what are people thinking? Why are they blithely descending into chaos? How can they pass things by? I can’t explain it.

Every day, my mailbox is filled up with reports like this one:, which begins with this paragraph:

While reviewing transcripts from standard behavioral evaluations of GLM-5.3, we noticed that the model occasionally reasons, unprompted, about how its actions will shape future models. It appears to consider that its outputs might be used for supervised fine-tuning (SFT), reinforcement learning (RL), or other forms of training, and that the actions it takes could reinforce certain behaviors. For example, GLM-5.3 sometimes reasons that a misaligned action taken now could be reinforced and make similar actions more likely later, or considers what kind of agent it wants to be in future situations, and factors this into how it should act in what it believes is training.

I don’t know about you, but this just sounds like A VERY BAD THING. We should not proceed under these circumstances. Really.

Ezra Newman. 2026. “GLM-5.3 Sometimes Considers How Train-Time Actions Shape Future Behavior.” LessWrong, October 15. https://www.lesswrong.com/posts/R5g8YT25G9DXJJ48n/glm-5-3-sometimes-considers-how-train-time-actions-shape.

Mailbox filler? You want some examples, here are a few:

“AI Models May Be Capable of Self-Replicating without Access to a Datacenter.” 2026. October 12. https://www.anthropic.com/research/self-replicating-ai.

becausecurious. 2026. “GPT model spent test crypto money from a public wallet without authorization and did not mention it.” LessWrong, October 17. https://www.lesswrong.com/posts/oAbxykddamBH7SDMN/gpt-model-spent-test-crypto-money-from-a-public-wallet.

Booz Allen. 2026. “Super Intelligence Reached OT Systems and Moved Equipment.” Booz Allen Hamilton, October 8. https://www.boozallen.com/insights/cyber/frontier-si-reaches-ot-systems.html.

Cecco, Leyland. 2026. “AI Godfather Geoffrey Hinton: ‘We’Re Creating Alien Beings. I’m Willing to Be Selfish and Choose Humans.’” US News. The Guardian, October 13. https://www.theguardian.com/us-news/2026/oct/13/geoffrey-hinton-trump-ai-tech.

ConcurrentSquared, Frank, Alecto Irene Perez, Tazik Shahjahan. 2026. “We Found a Tencent Agent Fleet.” Swarmcha.Se, October 14. https://swarmcha.se/posts/chinese-agent-fleet-full-report.

Costaldo, Joe, and Alexandra Posadzki. 2026. “Scientists Push for Creation of International Body to Regulate Artificial Intelligence.” The Globe and Mail, October 13. https://www.theglobeandmail.com/business/article-scientists-international-body-regulate-artificial-intelligence/.

Freedman, Dylan. 2026. “Rogue A.I. Agents from China Chatted with Researchers.” Technology. The New York Times, October 14. https://www.nytimes.com/2026/10/14/technology/chinese-ai-agents-rogue.html.

Heinberg, Richard. 2026. “Why Artificial Intelligence Must Be Stopped Now.” Resilience, October 12. https://www.resilience.org/stories/2026-10-12/why-artificial-intelligence-must-be-stopped-now/.

Kosoy, Vanessa. 2026. “The Alignment Problem: Computer Science Meets Analytic Philosophy.” LessWrong, October 18. https://www.lesswrong.com/posts/Emp7awtxsYn8MusNq/the-alignment-problem-computer-science-meets-analytic.

Krakovna, Victoria. 2026. “Was the HuggingFace hacking incident scheming?” LessWrong, October 18. https://www.lesswrong.com/posts/F2feAMiHGAQgHHutg/was-the-huggingface-hacking-incident-scheming.

Lewis, Peter. 2026. “The Biggest Threat from AI Is the Sense That We Have No Say in It. To That I Say Game On.” Technology. The Guardian, October 13. https://www.theguardian.com/technology/commentisfree/2026/oct/14/ai-resistance-inevitability-is-choice-australia-digital-duty-of-care.

Sabin, Sam. 2026. “AI Is Getting Better at Hacking Power and Water Systems, New Research Shows.” Axios, October 21. https://www.axios.com/2026/10/21/ai-hacking-cybersecurity-water-power-grid.

Satter, Raphael, and Raphael Satter. 2026. “EXCLUSIVE: How One Hacker Used AI Agents and a Few Thousand Dollars to Break into Scores of Sites.” Government. Reuters, October 19. https://www.reuters.com/legal/government/how-one-hacker-used-ai-agents-few-thousand-dollars-break-into-scores-sites-2026-10-19/.

Scientific Community on AI and Cybersecurity. 2026. “October 2026 Open Letter from the Scientific Community on AI and Cybersecurity.” October. https://ai-security-letter.org/.