AI Regulation & Policy

Today's Stories

1. 1

   A former cinematographer is now building governance and safety systems for AI agents that can act autonomously on behalf of humans.

   Someone who spent 30 years filming movies is now working on safety systems for AI agents—autonomous software that can take actions on a user's behalf. The shift reflects a broader move by experienced technologists into AI governance and control mechanisms. As AI systems become more capable and independent, the ability to oversee and constrain what they do grows more urgent. Bringing expertise from other fields—in this case, filmmaking and visual control—into AI safety suggests the challenge is broad enough to need diverse talent, not just traditional AI researchers.

   The piece signals a career transition rather than a product launch or specific timeline, but it illustrates how serious builders view the governance problem: important enough to pull talent away from established industries and toward making AI systems safer and more controllable.

2. 2

   Holden Karnofsky acknowledges that AI safety work carries genuine downside risks and may ultimately cause harm rather than prevent it.

   Karnofsky published a list of potential negative consequences of AI safety interventions, stating he is 'not aware of a good list of downside risks for AI safety broadly' and decided to make one. He notes that AI governance interventions are 'obviously high-variance' and that bad regulation can easily worsen outcomes. Karnofsky, a prominent figure in AI safety funding and strategy, is publicly articulating serious doubts about whether safety efforts net-positive — he describes living with 'the possibility that my ultimate impact on the utilons or whatever is going to be negative.' This challenges the assumption that safety work always reduces risk and signals that even dedicated practitioners must grapple with unintended consequences.

   Karnofsky frames this as a personal intellectual exercise ('the ones that I personally take seriously') rather than a comprehensive catalogue, and explicitly acknowledges the list is 'not intended to be fully comprehensive.' The piece suggests ongoing reflection within the safety community about whether interventions might sometimes backfire.

3. 3

   LessWrong essay argues that the most impactful AI governance work happens invisibly inside government and international institutions, not in public campaigns.

   An essay on LessWrong claims that strategic AI governance work splits into two categories — visible public work (press, open letters) and invisible insider work within ministerial cabinets and international institutions — and that much of the most consequential effort falls into the latter, largely unseen by the community. The AI governance community may be overinvesting in intellectual and public-facing work while undervaluing the insider roles that shape policy at the executive and international level. The author suggests this visibility bias means the community may not be allocating talent and attention where it would have the greatest effect.

   The author expresses hesitation about replicating the model of visible public campaigns (citing ControlAI in France as an example) in other countries, suggesting that understanding which governance strategies are truly invisible and impactful could reshape how the AI community approaches policy influence.

4. 4

   Amazon opposes AI systems that require human review at every step, arguing such safeguards slow down deployment and give competitors an edge.

   Amazon has publicly criticized "human-in-the-loop" AI governance—the practice of having people review and approve AI decisions before they take effect. The company views this approach as a barrier to rolling out AI systems quickly. Amazon's stance reflects a broader tension in the AI industry between speed and caution. By opposing mandatory human oversight, Amazon is signaling that it prioritizes rapid deployment over the kind of staged, supervised rollout that regulators and safety advocates favor. This preference could influence how other major tech companies approach AI governance.

   The outcome of this debate will shape whether AI systems launched by large companies include built-in human checkpoints or move to full automation. Amazon's public opposition may foreshadow industry pushback against tighter human-review requirements.

5. 5

   IBM and ServiceNow are positioning themselves as the primary vendors for governing AI agents—autonomous systems that can make decisions and take actions—as enterprises look to manage risks from these increasingly complex tools.

   IBM and ServiceNow have both introduced governance solutions aimed at controlling and monitoring AI agents, which are AI systems designed to operate autonomously and make decisions without constant human intervention. Both vendors are framing these offerings as essential tools for enterprises deploying such systems at scale. As companies adopt AI agents to automate business processes, they face new compliance, security, and accountability challenges that traditional AI governance frameworks may not fully address. IBM and ServiceNow are competing to establish themselves as the trusted vendors that enterprises rely on to keep these systems safe, transparent, and aligned with business rules.

   The competition between IBM and ServiceNow reflects a broader shift in enterprise software toward autonomous AI systems. Whichever vendor succeeds in becoming the standard for agentic AI governance could capture significant mindshare and adoption among large enterprises that are building their AI automation strategies.

6. 6

   EU retailers want AI-generated marketing content exempt from deepfake rules, but the bloc hasn't defined what a deepfake actually is.

   Eurocommerce, the trade association representing Amazon, H&M, and IKEA, is pushing for AI-generated ads to be exempted from the EU AI Act's transparency requirements. The group argues that AI-generated product images—such as a living room scene used to sell a sofa—do not qualify as deepfakes. Zalando reports that 90 percent of the marketing content on its platform is already AI-generated. The EU has not established a clear definition of what constitutes a deepfake, creating ambiguity about which AI-generated content falls under the AI Act's rules. Retailers face uncertainty about compliance, since a large portion of their marketing content is now AI-generated but may or may not be subject to disclosure rules depending on how the regulation is eventually interpreted.

   How EU regulators define the boundary between deepfakes and permissible AI-generated marketing. The outcome will determine whether retailers must add transparency labels to the majority of their platform content or whether a broad exemption for commercial imagery takes hold.