AI Regulation & Policy

Today's Stories

1. 1

   Safety researchers ask: can open-weight AI model safety survive rapid fine-tuning attacks?

   A machine learning researcher raised a question about the practical value of safety training in open-weight language models (AI systems that understand and generate text), noting that modified versions removing safety guardrails appear within 30 minutes of a model's release using automated scripts. The question challenges whether the cost and effort of safety training is justified if determined users can quickly disable those protections through fine-tuning (retraining) or by switching to alternative models. It raises a core tension for companies releasing open models: building meaningful safeguards when motivated actors can circumvent them anyway.

   The researcher is asking the community how to define meaningful defensive wins—such as raising attacker costs or making safety removal less reliable—even if perfect prevention is impossible, signaling this is an open debate within the AI safety field rather than a settled question.

2. 2

   Intuit CEO frames AI governance through U.S. Constitution lens

   Intuit CEO Brad Smith published a commentary drawing parallels between the U.S. Constitution's founding principles and the challenges of governing artificial intelligence today. He emphasizes that the framers created a system balancing stability with adaptability—distributing authority across competing institutions—and argues similar principles should guide AI governance. As businesses and policymakers grapple with AI's risks and benefits, Smith suggests the Constitution offers a blueprint: protect individual rights while fostering innovation, establish standards for AI safety, invest in public-private partnerships for AI education, and ensure diverse representation in training data and evaluation frameworks. For business leaders, the message is that durable systems require both organizational agility and cross-sector collaboration rather than isolated corporate innovation.

   Smith highlights that Intuit has practiced this adaptability for 40+ years, disrupting itself across multiple technology eras—from DOS disks to the web, mobile, cloud, and now AI. He calls for a "barn raising" mentality: collaborative governance rooted across companies, academia, government, and everyday people to ensure AI development is safe, ethical, and inclusive.

3. 3

   Agentic AI Security Market to Hit $13.52B by 2032 on 42% Annual Growth

   The agentic AI security market is projected to expand from USD 1.65 billion in 2026 to USD 13.52 billion by 2032, with a compound annual growth rate of 42.0%. This surge is driven by rising use of third-party AI tools that need secure integration within AI agent environments. Financial services and banking firms are expected to lead adoption, as they deploy autonomous AI systems for fraud detection and risk management. These sectors face substantial security risks when handling sensitive financial data, making robust security measures essential to prevent unauthorized transactions and data breaches.

   AI governance and risk platforms are projected to see the highest growth rate during the forecast period. These platforms enable enterprises to maintain oversight, transparency, and compliance as they deploy multi-agent systems, helping mitigate risks like decision drift and unauthorized data use.

4. 4

   U.S. lifts AI export controls on Anthropic's Fable model

   The U.S. government reversed export controls it had imposed two weeks earlier on Anthropic's Mythos and Fable models. The government first lifted controls on Mythos on Friday evening, then on Fable late Tuesday. Both models had been disabled for all users during the restriction period. The temporary controls exposed a risk that U.S. businesses now recognize — they cannot reliably depend on American frontier AI models for essential tasks without fallback options. This concern is pushing more enterprises to explore open source alternatives, even though the most capable open source models currently come from Chinese AI companies, creating a dilemma over reputational and geopolitical risks.

   The U.S. is working with leading AI labs on explicit voluntary cybersecurity standards that frontier AI models can meet to avoid government objection to public release. Anthropic announced it is also developing a shared framework with Amazon, Microsoft, Google, and other critical infrastructure partners to assess risks from jailbreaks to model guardrails.

5. 5

   Warp raises $60M to automate payroll and HR with AI

   Employee management startup Warp announced $60 million（約96億円） in new funding. The New York-based company is positioning itself as an AI-native alternative to legacy human capital management software, using artificial intelligence to handle payroll and back-office work with minimal staff involvement. The human capital management category has long been dominated by established players like Workday. Warp's approach suggests a shift toward AI-driven automation for roles traditionally requiring dedicated HR and payroll teams, which could reshape how businesses manage these functions.

   The company is pitching itself as an alternative to the incumbents that have controlled this market. The funding signals investor confidence in AI-native approaches to enterprise HR operations, though real-world adoption and competitive response from established vendors remain to be seen.

6. 6

   Enterprise AI control gap widens as platforms multiply, ownership unclear

   A VentureBeat Pulse Research study found that AI portfolios are expanding faster than enterprises can govern them. Most organizations operate multiple competing AI platforms, struggle to detect model failures in production, and lack a single owner accountable for AI across their technology stack. The widening gap between spending and visibility creates real financial and operational failures, with autonomous agents already producing measurable losses. For business leaders, this means spending on AI without the ability to see what is actually working, who is responsible when things break, or what the true costs are.

   The research identifies the absence of clear AI ownership as the single most-cited barrier to cross-platform governance. Organizations currently govern AI control mostly by hand, suggesting those who establish clear accountability and visibility structures may gain a significant operational advantage.