Anthropic AI Updates: July 29, 2026

1. Claude Discovers Novel Cryptographic Weaknesses

Anthropic. Anthropic’s Frontier Red Team reported that a Claude preview model uncovered mathematical flaws in cryptographic algorithms themselves rather than mere implementation bugs, improving an attack on the HAWK post-quantum signature candidate that effectively halved its key strength and using a new “Mobius Bridge” fingerprinting technique to speed attacks on 7-round AES by 200 to 800 times. The company stressed that none of the findings affect production systems today, but they show AI models can now perform cryptanalysis at expert levels. Anthropic warned that many ciphers protecting modern systems have received less scrutiny than they deserve and may harbor dormant weaknesses, framing the work as a tool for stress-testing and developing new cryptographic standards. Source

2. Anthropic Ships MCP 2026-07-28 With a Stateless Core

Anthropic. Anthropic released MCP 2026-07-28, the fifth specification version of the Model Context Protocol, shifting the protocol from a bidirectional stateful design to a request/response model that can run on serverless and edge infrastructure. The update introduces a versioned extensions framework so capabilities like MCP Apps and Tasks can add interactive UIs and long-running operations without changing the core protocol, and tightens authorization with updated OAuth 2.0 and OIDC alignment for enterprise identity systems such as Entra and Okta. Anthropic said Claude’s directory now hosts more than 950 MCP servers and cited over 400 million monthly SDK downloads, a fourfold increase, positioning the changes to simplify building and scaling production integrations. Source