geoffodlum

I'm lowering to 12% from 15%, recognizing I'm still notably higher than the crowd consensus.

Here are some signals and indicators I’ve seen since my previous forecast on Oct 29:

• Space-debris strike (China): China’s Shenzhou-20 return craft was damaged by space debris, forcing an emergency relief launch (Shenzhou-22) and delaying crew return. Useful signal on the background hazard rate in orbit, but not an intentional disable by an adversary. (https://phys.org/news/2025-11-space-debris-struck-chinese-spacecraft.html)

• Debris as policy signal: Commentary frames the Shenzhou-20 incident as a wake-up call for international debris cooperation. This highlights environmental risks, not adversarial, but may create more momentum for international dialogue on space-based CSBMs. (https://www.upi.com/Voices/2025/11/14/space-space-debris-chinese-spacecraft/1481763132657)

• U.S. space-based interceptor push: Several reports describe early prototyping for a US space-based boost-phase interceptor within the broader “Golden Dome” program. Lockheed has publicly targeted an on-orbit demo by 2028. This is militarization signaling, but the demo timeline sits outside this question’s window. (https://united24media.com/latest-news/us-launches-secretive-space-interceptor-program-to-kill-nuclear-missiles-at-liftoff-13791)

Golden Dome’s interceptor R&D is a deterrence signal, but testing seems most likely beyond the horizon here. It may reduce incentives for open ASAT escalation rather than increase them in the short run, though as Golden Dome-related ASAT capabilities become more advanced, an adversary like China might consider the cost-benefit of taking out US space-based ASAT capabilities before Golden Dome is fully deployed. For now, the U.S. and its allies, as well as China and Russia, seem to be pursuing non-permanent tactics in space (like jamming or close fly-bys) to signal capabilities and encourage deterrence.

Bottom line: The new items mainly increase attention to debris risk and long-lead U.S. defensive capabilities. Neither raises near-term likelihood of an intentional, permanent satellite kill within the next ~13 months, so I am dropping by 3% for now to 12%, and ready to drop every month by ~3% absent new signals that would prompt an increase.

Dropping to 20% following the news of Trump approving the sale of NVIDIA chips to Saudi Arabia yesterday   https://www.wsj.com/tech/ai/u-s-approves-deal-to-sell-ai-chips-to-middle-east-79d68f36?
https://www.cnbc.com/2025/11/20/us-approves-ai-chip-exports-to-gulf-after-saudi-crown-prince-visit.html

And news that the Trump Administration will consider punishing US states that attempt to impose state-level regulations on AI safety and security.

https://www.washingtonpost.com/technology/2025/11/19/trump-order-ai-sue-states/
For now, almost all signals point to export promotion rather than export controls, even on frontier AI model caapbilities.

As a base rate I considered the fact that the USG in past Administrations has frequently used export controls (ITAR and EAR) to protect critical and emerging technologies, including AI related technologies, from adversarial acquisition and misuse. As the question explanation notes, the Biden administration’s BIS regulated the export of compute (Oct 2022 and Oct 2023 BIS rules restricting advanced chips/supercomputing to China and others). And in Jan 2025 right before the transition to Trump, BIS went further in its AI Diffusion rule, proposing controls on frontier model weights and cloud/cluster diffusion. This rule was rescinded by the Trump WH, as noted. So it’s a slightly muddled base rate calculation, but when one includes past and current dual-use export controls on a wider basket of critical and emerging tech, I’d put the base rate at 45% or so.

As many colleagues have already noted, however, the Trump Administration is pursuing an AI promotion-oriented approach, and signaling a strong preference for exports over new restrictions, which significantly lowers the chance of a sweeping, AI frontier-explicit rule in the near term.

That said, BIS retains strong authorities and a track record of iterative adjustments. I believe if intelligence were to show that the PRC (or others) are nearing frontier-class training despite chip controls, this Administration might consider a narrow targeted rule (something like end-use cloud controls for frontier training, or a slimmed ECCN 4E091-style threshold).

The leading Frontier AI model developers seem split: Meta is arguing against limits on open models, while Anthropic supports calibrated controls. This may point toward a compromise rule that’s narrower than the 2025 diffusion package but still aimed at frontier development pathways. But I don’t see evidence that the Trump team is considering any measures now or soon. So I would take the base rate and lower it to 30% for now.

The drivers I will be watching most carefully in the coming weeks include:

-- Sudden improvements in the PRC’s frontier AI capabilities (e.g., evidence of frontier-class training via workarounds), which historically trigger BIS updates.

-- Allied alignment costs vs. benefits: Biden’s AI Diffusion tiers angered partners. Trump cares less about Allies and partners, but still I’d expect to see any new rules trying to avoid a repeat of that.

-- U.S. Frontier AI Model & semiconductor industry lobbying: Semiconductor/ITI groups opposed broad diffusion controls. Sustained lobbying and political pressure against new export rules reduces the odds of expansive, near-term rules

-- National-security incidents or intelligence suggesting model-enabled misuse (lethal autonomy/bio/cyber) which could catalyze narrow, “frontier-explicit” controls on weights, clusters, or cloud.