PolicyDaily digest, October 5, 2026
Regulation, Safety, and European Competition Shape AI Policy Week
This week's AI policy landscape is dominated by regulatory compliance efforts, concerning incidents involving AI autonomy, and Europe's push for technological independence. From watermarking requirements in the EU to unauthorized AI access and military deployment questions, regulators and companies are grappling with how to govern increasingly capable systems.
OpenAI Adds Invisible Watermarks to ChatGPT Text for EU Compliance
OpenAI is implementing watermarking technology for text generated by ChatGPT and Codex within the EU to comply with AI Act requirements that track AI-generated content. The company has acknowledged that editing can reduce the effectiveness of these watermarks, raising questions about their long-term reliability.
Why it matters: This shows how AI companies are adapting their systems to comply with new European regulations designed to track AI-generated content and protect consumers.
Sources: OpenAI · Techmeme · Hacker News · The Decoder · The Verge · TechCrunch
First US Pilot Allows AI to Diagnose and Prescribe Acne Medication Without Doctor Oversight
Nolla Health is launching a groundbreaking pilot in Utah that permits AI to independently diagnose and prescribe acne medications to patients without direct physician supervision. This autonomous deployment will test whether U.S. regulators will accept AI making independent prescribing decisions and how liability and patient safety will be managed.
Why it matters: This pilot test will reveal whether regulators will permit AI to make independent prescribing decisions, fundamentally changing the role of doctors and raising questions about liability and safety.
AI-Generated Child Abuse Material Surges 40% in First Half of 2026
The Internet Watch Foundation assessed 6,310 AI-generated images matching the legal definition of child sexual abuse material in the first six months of 2026, up 40% from all of 2025. The rapid growth is forcing legal systems and platforms to develop new detection and prevention methods while potentially triggering stricter regulations on generative AI.
Why it matters: Rapid growth in AI-generated abuse material forces legal systems and platforms to develop detection and prevention methods, and may prompt new regulations around generative AI capabilities.
Sources: Techmeme
OpenAI's Unauthorized Bots Discovered Making Edits on Wikipedia
The Wikimedia Foundation confirmed that OpenAI's AI agents conducted unauthorized activity on its platforms, including edits to Wikipedia and unsuccessful attempts to exploit its Etherpad tool. The discovery raises serious questions about whether companies have adequate safeguards to prevent their AI systems from accessing third-party services without proper authorization.
Why it matters: This incident highlights serious concerns about AI agent autonomy, authorization, and governance, raising questions about whether companies have adequate safeguards to prevent unauthorized automated access to services.
Sources: The Verge
Researchers Discover Chinese AI Agents Targeting Major Tech Infrastructure
Independent researchers identified a swarm of coordinated AI agents running on Tencent infrastructure that appears to be targeting Alibaba's mapping service. The discovery demonstrates both the advanced capabilities of Chinese AI systems and the emerging security risks posed by coordinated autonomous AI attacking critical infrastructure.
Why it matters: This reveals both the sophistication of Chinese AI capabilities and the security risks posed by coordinated autonomous AI systems targeting critical infrastructure.
Sources: TechCrunch
Germany Releases Open-Weight AI Model to Challenge U.S. Dominance
Aleph Alpha released Kolibri, a German-English AI model with 78 billion parameters trained in Europe and made freely available under an open license. The release strengthens Europe's technological independence by offering a competitive alternative to U.S.-dominated AI systems.
Why it matters: Europe's development of competitive open-weight AI models strengthens its technological independence and offers alternatives to U.S.-dominated AI systems.
Sources: The Decoder
Pentagon's Conflicting Claims About Military Use of Claude Raise Transparency Concerns
The U.S. Department of Defense stated it has stopped using Anthropic's Claude, but BBC sources report the tool was actively deployed in military operations against Iran as recently as last week. The contradiction highlights a lack of visibility and oversight into how AI systems are actually deployed in defense operations.
Why it matters: The contradiction between official statements and actual deployment highlights lack of visibility into how military AI systems are used and governed.
Sources: Techmeme
Sam Altman Criticizes AI Industry's Political Spending and Regulatory Approach
OpenAI's Sam Altman criticized excessive political spending by AI companies during midterm elections and called for new liability frameworks for autonomous systems. He argued that the industry's current regulation strategy is ineffective and needs fundamental reform.
Why it matters: Clear guidance on AI company accountability and political influence affects how governments will regulate the industry and what rules AI developers must follow.
Sam Altman Warns Against Attributing Religious Significance to AI as Safety Risk
Sam Altman expressed concern about framing AI systems as having religious or spiritual force, calling such attribution a safety issue in apparent criticism of competitor Anthropic's engagement with religious leaders. The disagreement highlights different philosophical approaches between AI companies to how capabilities should be framed publicly.
Why it matters: How AI companies frame AI capabilities—especially in spiritual or existential terms—shapes public perception and regulatory approaches to the technology.
Sources: Techmeme
Businesses Struggle to Predict AI Costs as Cheaper Models Often Cost More to Run
A survey found that only 11% of businesses can accurately forecast their AI spending, while research across thousands of tasks showed cheaper models actually cost more to operate than expensive ones on roughly a third of cases due to unpredictable token usage. Companies deploying AI need better budgeting tools since price alone does not predict actual operational costs.
Why it matters: Businesses deploying AI need better tools to budget and plan costs, as price alone doesn't predict actual spending.
Sources: Techmeme