Visa integrated payment functionality directly into ChatGPT, allowing AI agents to make purchases at any Visa merchant without user intervention for each transaction. The integration marks a shift toward autonomous AI systems handling financial decisions independently, though users retain control over spending limits and merchant categories.
The move reflects broader momentum in autonomous AI this week. Anthropic released Claude Fable 5, its most powerful publicly available model, to all users. Jeff Bezos simultaneously emerged from stealth with Prometheus, a startup capitalized at $41 billion to build what he frames as an "artificial general engineer" focused on automating complex technical work.
Security concerns surfaced when researchers discovered a self-replicating worm compromising 73 of Microsoft's internal GitHub repositories through AI coding tools. The vulnerability highlights risks as AI systems gain deeper access to infrastructure and decision-making systems.
Regulatory friction mounted on multiple fronts. Anthropic publicly disagreed with the White House over federal preemption of state AI laws, arguing states should retain autonomy in setting guardrails. Meanwhile, a German court ruled Google liable for factual errors in its AI Overviews search results, establishing legal liability for AI-generated content accuracy.
The week underscores a critical tension in AI development. Capabilities are expanding rapidly. Visa's payment integration, Claude Fable 5's public rollout, and Prometheus's infrastructure-building focus all demonstrate genuine progress in autonomous decision-making and reasoning. Yet each advancement increases systemic risk. An AI shopping agent spending unchecked, worms replicating through code repositories, and search systems spreading misinformation without clear liability create new vulnerabilities.
The regulatory responses from Germany and Anthropic suggest the industry is fragmenting around liability and safety standards. Vendors push autonomy forward. Courts and policy makers scramble to establish who bears responsibility when autonomous systems fail. This pattern will