Anthropic tops OpenAI in adoption

Anthropic moved ahead of OpenAI in one closely watched measure of paid business adoption just as House Homeland Security lawmakers were briefed behind closed doors on the company’s cybersecurity-focused Mythos model. Ramp said on May 13 that 34.4% of businesses in its sample paid for Anthropic services in April, compared with 32.3% for OpenAI. On the same day, the House Committee on Homeland Security held a closed-door session on Mythos, according to people familiar with the meeting cited by The Hill and CyberScoop. Ramp’s dataset covers more than 50,000 American businesses using its corporate card and bill-pay platform, while Anthropic has described Mythos as a model with unusually strong computer-security capabilities. ### How big was Anthropic’s lead in Ramp’s data? Ramp said Anthropic’s adoption rate rose 3.8% in April to 34.4% of businesses, while OpenAI’s fell 2.9% to 32.3%. The company said it was the first time Anthropic had passed OpenAI in its monthly AI Index. Overall paid AI adoption across Ramp’s sample rose 0.2 percentage points to 50.6%, Ramp said. Ara Kharazian, Ramp’s lead economist, said the index is based on transaction data rather than surveys. (ramp.com) Ramp says it identifies adoption when a business records a positive transaction amount for an AI product or service in a given month, using merchant names and line-item details from receipts and bills. The company also says the measure likely understates actual use because it does not capture free tools or employee purchases made on personal accounts. ### What exactly does Ramp measure — and what does it miss? Ramp says its AI Index tracks paid adoption among American businesses on its platform, not total market share across all enterprises. The sample includes more than 50,000 businesses and “billions of dollars in corporate spend,” according to the methodology page. That makes the index a measure of verified spending behavior inside Ramp’s customer base, rather than a census of all corporate AI use. (ramp.com) May 13’s Ramp note also cautioned against treating one month of data as a final verdict on the market. Kharazian wrote that the results “should not be construed” as proof Anthropic is the definitive leader, and he pointed to pricing, compute constraints and rising use of cheaper open-source models as risks to Anthropic’s position. ### Why were lawmakers meeting Anthropic behind closed doors? (ramp.com) The House Committee on Homeland Security hosted Anthropic for a closed-door briefing on Wednesday focused on Mythos, The Hill reported, citing three sources familiar with the meeting. Those sources said the discussion would cover Mythos’s capabilities, national-security implications and policy considerations. The Hill reported the Anthropic side was led by Logan Graham from the frontier red team and Josh Tilstra from national security programs and policy. (ramp.com) CyberScoop reported the May 13 session included a live demonstration of Mythos for lawmakers and staff. A committee aide told the publication the demo showed how advanced AI could identify and reason through software vulnerabilities, and CyberScoop said lawmakers are considering more oversight steps, including a hearing planned by cyber subcommittee chairman Andy Ogles and a request from committee Democrats for a classified briefing. (thehill.com) ### What has Anthropic said Mythos can do? Anthropic said on April 7 that it launched Claude Mythos Preview and Project Glasswing to test and deploy the model in cybersecurity work. In a technical post, the company said Mythos was capable of identifying and exploiting zero-day vulnerabilities in major operating systems and web browsers when directed by a user, and said more than 99% of the vulnerabilities it found had not yet been patched. (cyberscoop.com) Anthropic also said the oldest bug it had found so far was a now-patched 27-year-old flaw in OpenBSD. April 7’s Anthropic post said the company was limiting disclosure of many findings because of coordinated vulnerability disclosure concerns. The company framed Project Glasswing as an effort to help secure critical software and prepare defenders for more capable security models. ### What happens next in Washington and the market? Andy Ogles, the Republican chair of the Homeland Security cyber subcommittee, told CyberScoop he plans to hold a hearing related to Mythos after the May 13 briefing. (red.anthropic.com) Committee Democrats are also seeking a classified briefing, according to CyberScoop. On the business side, Ramp will publish future monthly AI Index updates from the same transaction-based dataset, which it says tracks paid AI adoption among U.S. businesses on its platform. (cyberscoop.com)