Week 42

Huang signed and told reporters, per CBS, There's no conflict between innovation, technology and safety. It fits his usual line and he said nothing new on risk. Scores hold at 5 and 5.

Musk signed the Joint Commitment on Frontier Responsibilities on September 29 and explained it in his own terms: We agreed to a number of things that include joint monitoring, board special committees, just generally grading each other's homework. In a second account he added that this is a lot better than if people just grade their own homework. The commitment is voluntary and he offered no view on whether it will bind anyone. Scores hold at 5 and 4: he is backing outside checking, but still as an industry arrangement.

Zuckerberg signed on September 29. The Korea Times reports he said the accord is meant to show AI works in the way that we intend and called it a starting point for industry-wide standards. That is a customer-assurance framing more than a risk framing. Scores hold.

On October 3 Altman posted on X: I am very uncomfortable about people trying to ascribe religious force or a surrender of human judgment to AI models, and think it is a real safety issue. Coverage ties it to a New York Times report that Anthropic's Chris Olah has met religious leaders about model consciousness and morals. An Anthropic spokesperson said the talks were not about giving Claude religious authority. The post is a safety argument aimed at a rival rather than a change in his own forecast, so scores hold at 5 and 3. OpenAI's president Greg Brockman signed the accord on the company's behalf; Altman himself is not quoted on it in anything we could verify.

Clark is a co-author of the September 28 paper. The Anthropic figures it cites, over 80% of approved code written by AI and R&D done under light supervision rising from 1% to 26%, are the company's own. Putting them in a public warning is consistent with his position. Scores hold at 5 and 3.

Amodei signed on September 29. The Korea Times quotes him: the technology has very real risks, and the mechanism, how we address those risks, is still under discussion, and whoever wins AI, wins. I think that's very important. The first line keeps his earlier position that the accord is a start. The second is a competitive framing in the same breath as the call for oversight. He did not repeat last month's call to slow down in anything we could verify this week. Scores hold at 5 and 2.

Bengio co-authored the September 28 intelligence-explosion paper (see Hinton). His position is unchanged: the paper's asks, embedded auditors, capability limits and international agreements, go further than the voluntary accord signed the next day. Scores hold at 5 and 2.

Hinton is a co-author, with Bengio, Clark, OpenAI's Jakub Pachocki and more than 20 others, of a paper on what happens if AI research automation triggers an intelligence explosion. It reports AI writing over 80% of approved code at Anthropic and light-supervision R&D rising from 1% to 26% between March and August 2026, and says Once an intelligence explosion begins, the window for action may close. It asks for visibility into lab R&D automation, embedded auditors, capability limits and international agreements. No separate Hinton statement was verified. Scores hold at 5 and 2.

Added as a guest. OpenAI's chief scientist is a co-author of the September 28 paper warning that once an intelligence explosion begins the window for action may close. It is rare for the person running OpenAI's research to put his name to that. No separate statement from him was verified, so the reading rests on the paper alone and the scores (5 and 2) are provisional.