Anthropic has announced another $20 million donation to Public First Action, a bipartisan advocacy group that works on AI transparency, safety and public education, the company said on Wednesday. With this new donation, Anthropic's total funding to Public First Action has reached $40 million, according to the Anthropic announcement.Anthropic donates $40 million for AI safety and policy as Claude Mythos raises security concerns (REUTERS/Dado Ruvic/Illustration/File Photo) (REUTERS)The company said it decided to double its funding because AI models are becoming much more powerful, including its own Claude Mythos model, and this creates bigger risks if they are misused. Anthropic said the money is meant only for Public First Action's public education and AI policy work, and cannot be used to help or oppose any federal, state or local election candidate.Why Anthropic gave more money for AI safetyPublic First Action is the policy arm of Public First and is linked to super PACs that support candidates from both major political parties who back stronger AI safety rules. Anthropic said its donation is one way to make the public and lawmakers pay more attention to the urgent debate over AI regulation and safety. The company had already donated $20 million to Public First Action in February, making this its second major contribution.Claude Mythos and AI security risksAnthropic said the timing of the latest donation is linked to the release of Claude Mythos Preview, which it described as its most advanced AI cybersecurity model at the time, as noted by The Hill. The Claude Mythos Preview model was first made available only to selected cybersecurity, technology, financial institutions and government users. Also read: From Michigan to Mumbai, communities are asking: How much water do data centres use?Anthropic warned that if highly capable AI models fall into the wrong hands, they could threaten critical infrastructure such as hospitals, power grids and other essential systems. The company also said advanced AI can bring major benefits in science, medicine and the economy, but strong safeguards must be in place before those benefits can be fully realised. Anthropic's latest donation comes as AI companies and their executives are becoming much more active in US election-related policy and advocacy efforts ahead of the midterm elections.AI policy debate grows before electionsRecent campaign finance filings showed Anthropic CEO Dario Amodei personally donated $1 million to the same super PAC, while several Anthropic employees together contributed another $2.15 million during the last quarter, according to The Hill. Across the AI industry, advocacy groups now have large amounts of money available as AI becomes a bigger political issue during the election season.One such group is Leading the Future, a super PAC backed by technology executives and investors that supports faster AI development with lighter government regulation. Anthropic has supported transparency rules for AI, but it wants governments to go even further by introducing stronger safety requirements before powerful AI systems are released.Dario Amodei wants tougher AI rulesLast month, CEO Dario Amodei proposed that governments should have the authority to stop the release of dangerous AI systems if they fail to meet required safety standards, as stated by The Hill. Amodei also called for mandatory independent testing by third-party auditors to check AI models for risks such as cyberattacks, biological weapons, loss of human control and dangerous automated research capabilities.Some of Anthropic's competitors disagree with these proposals, saying stricter rules could slow down AI development in the US and reduce its ability to compete with China. Anthropic acknowledged growing national security concerns and said the US and its democratic allies currently lead in AI, but warned that this advantage could be lost without stronger policies.The company also urged tighter export controls on advanced AI chips and semiconductor manufacturing equipment, along with policies to stop foreign groups from copying or distilling advanced US AI models.