Anthropic tightens access and adds a healthcare voice

Anthropic is tightening who gets its most powerful model and who helps govern the company. On April 14, it added Novartis chief executive Vas Narasimhan to its board after withholding Claude Mythos Preview from public release last week. (anthropic.com 1) (anthropic.com 2) Anthropic said April 7 that Claude Mythos Preview is its “most capable frontier model to date,” but it did not launch the model for general use. Instead, the company published a system card and limited access through Project Glasswing, a security program for critical software defenders. (anthropic.com 1) (anthropic.com 2) Project Glasswing launched with Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks, and Anthropic said it also extended access to more than 40 additional organizations that build or maintain critical software infrastructure. (anthropic.com 1) (anthropic.com 2) The company’s cybersecurity team said the model is unusually strong at finding and exploiting software flaws, and said the release decision was shaped by tests on offensive cyber capabilities. Anthropic’s technical write-up said the goal was to give defenders “a head start” rather than publish the model broadly. (red.anthropic.com) (anthropic.com) That release strategy lands in the middle of a wider fight over frontier artificial intelligence, the newest class of large models trained with enough computing power to write code, analyze systems, and act more like autonomous workers. Anthropic’s own Responsible Scaling Policy, updated in February and again on April 2, says the company will tie deployment to capability and safeguard assessments. (anthropic.com) (anthropic.com) (anthropic.com) Outside coverage has treated the Mythos decision as a test of whether leading labs will keep top systems behind enterprise or controlled-access walls instead of releasing them widely. TechCrunch reported criticism from some developers who argued that gating powerful models could concentrate access among large companies. (techcrunch.com) Anthropic paired that access decision with a governance change. The company said Narasimhan was appointed by its Long-Term Benefit Trust, an internal governance body created to keep Anthropic aligned with its public-benefit mission rather than only investor returns. (anthropic.com) (corpgov.law.harvard.edu) Narasimhan is a physician-scientist and has led Novartis as chief executive since 2018. Anthropic said he brings experience in medicine, drug development, and global health as the company pushes further into healthcare and life sciences uses for artificial intelligence. (anthropic.com) (novartis.com) Anthropic’s board has been in flux for months. Techmeme, citing reporting from The Information, said Narasimhan is the company’s second board addition in recent months as Anthropic weighs an initial public offering and expands in healthcare. (techmeme.com) Taken together, the moves put Anthropic’s April strategy in plain view: keep Claude Mythos Preview inside a narrow release channel, and add a director whose background is in one of the industries the company says artificial intelligence could reshape first. (anthropic.com) (anthropic.com)