AI Leaders Push Independent Oversight as Risks, Lobbying Rise

Fears that artificial intelligence could enable devastating bioweapons or slip beyond human control have intensified in recent weeks, even as leading companies pour tens of millions into influencing the very regulations meant to address those dangers. Anthropic, the maker of the Claude chatbot, has funneled 40 million dollars through a super PAC to back candidates and shape congressional debates, prompting accusations from lawmakers that the effort serves to sideline smaller competitors under the banner of safety.

The spending comes amid stark warnings from inside the industry. A senior Anthropic researcher recently assessed a greater than 10 percent chance that AI could kill all humans within the next decade. Microsoft AI chief executive Mustafa Suleyman described controlling advanced systems as a really big challenge and warned that creating something more powerful than humanity combined would fail its fundamental purpose. Dario Amodei, Anthropic's own chief executive, has called for independent oversight bodies to evaluate frontier models before they advance further.

These concerns center on concrete pathways to harm. Experts highlight how AI tools could help far more people design and deploy synthetic pathogens by guiding gene editing and DNA synthesis, lowering the barrier for malicious actors without requiring superintelligence. At the same time, current AI capabilities are already accelerating a surge in software vulnerabilities. Microsoft alone issued patches for 974 common vulnerabilities and exposures this month, while Oracle shipped 1,448 in July alone. Researchers using Anthropic's models identified 271 flaws in Firefox during a single sprint, adding to a total of more than 66,000 CVEs recorded so far this year and straining security teams worldwide.

Anthropic has defended its regulatory push as necessary guardrails, yet critics including Senator Ron Johnson argue that true safety comes from broad competition rather than rules that could consolidate power among a handful of well-funded labs. The company's board includes major Democratic donors, and its affiliated PAC has directed millions into state races, including a 3 million dollar contribution in one New York contest. Rival firms have responded with their own spending, turning the debate into an expensive contest over who sets the terms for future oversight.

Developers report that models are already contributing substantially to their own improvement. Anthropic stated that Claude now leads 26 percent of its research and development work, completing complex tasks from high-level prompts under human supervision. This progress toward recursive self-improvement raises questions about how quickly systems could outpace the controls their creators claim to prioritize.

The immediate effects of AI-assisted discovery, from expanded cyber risks to potential bioweapon assistance, are already visible and demand resources that stretch far beyond voluntary corporate pledges. Lawmakers and public health experts have long called for greater investment in pandemic preparedness and open-source security maintenance, steps that would address both natural outbreaks and engineered threats without handing additional leverage to the firms racing ahead.