Anthropic unveiled 10 artificial intelligence agents designed exclusively for financial firms at a public event in New York on Wednesday, signaling the enterprise-focused AI company's aggressive push into the financial services industry. Dario Amodei, Anthropic's founder and CEO, also laid out a blueprint for the convergence of finance and AI in a fireside chat with Jamie Dimon, CEO of JPMorgan Chase, a leading figure in traditional finance.
The financial AI agents unveiled cover everyday tasks at financial firms, including automated Pitchbook creation for investment presentations, company valuations, discounted cash flow modeling, customer due diligence, month-end close automation, financial statement audits and anti-money laundering investigations. Most of these functions are already in use at financial firms. The agents can be accessed through Anthropic's Claude, as well as Microsoft's Excel, PowerPoint, Word and Outlook.
"Over the weekend, I opened Claude Code and asked it about asset swaps, U.S. Treasury spreads and market liquidity, and in 20 minutes it produced a massive dashboard and research. It was very accurate," Dimon said. He noted that JPMorgan has been integrating machine learning into finance since 2012 and currently operates hundreds of use cases, including risk management, fraud detection and document review.
Amodei predicted that AI models will become more accurate and reliable, and that autonomy will expand to the point where AI can carry out week-long projects from start to finish. He also hinted at Anthropic's next features, forecasting real-time integration of web and financial data.
Anthropic also disclosed its full-year revenue forecast, saying it would surpass $30 billion (about 41 trillion won). That is more than a three-fold jump from $9 billion last year.
"We had expected 10 times revenue growth, but annualizing the first quarter gave us 80 times," Amodei said. "It exceeded the maximum I could have imagined." He also expressed caution. "There is fundamental unpredictability in Anthropic's technology. As models improve, accuracy rises and autonomy grows, but no one knows how it will manifest in the real world," he stressed.
Amodei also addressed the cybersecurity debate surrounding Mithos. He disclosed for the first time that about 300 security vulnerabilities had been discovered in Firefox through Mithos. While Anthropic's Mithos has strong hacking detection capabilities in itself, there have been concerns about side effects if hackers exploit it.
"Chinese models are currently six to 12 months behind. If vulnerabilities are not fixed within that window, China will get its hands on the same tools. Conversely, if this time is used well, we could end up with an even more secure codebase," Amodei said.
