Avow is the SOC 2 platform we wished we'd had. We built it after our own SOC 2 Type 2, where the audit passed but the workflow was miserable, and where the incumbents wouldn't even show a price without booking a call. Evidence streams in real time from AWS, GitHub, and Google Workspace, you fix from an action feed instead of a dashboard maze, and your auditor gets a scoped, read-only portal. One framework, done deep. If you are heading into a SOC 2, or staring at a renewal hike from Vanta or Drata, take a look.

As for the week itself, the rest of the field kept moving. OpenAI previewed its GPT-5.6 family and expanded its Daybreak cybersecurity program, and the Model Context Protocol shipped its biggest spec revision since launch. The one headline from Anthropic worth noting: the US partially lifted its export ban on June 26, clearing Mythos 5 for about 100 vetted partners, though Fable 5 is still dark worldwide. Details below.

Claude Code this week

Claude Code shipped a run of releases since last issue - v2.1.186 through v2.1.193, the week of June 22. The highlights:

  • Shell auth for MCP servers: new claude mcp login and claude mcp logout authenticate a configured MCP server straight from your shell, instead of going through the interactive /mcp menu
  • Shell mode talks back: command output now gets an explanation without a second prompt - run npm test and Claude reads the result inline
  • /rewind reaches past /clear: you can now resume a conversation from before a /clear wiped it
  • Background subagents stop auto-denying: a permission prompt from a background subagent now surfaces in the main session instead of being silently denied
  • Smaller wins: live file-path autocomplete in bash mode; idle background shells get memory-reaped automatically (set CLAUDE_CODE_DISABLE_BG_SHELL_PRESSURE_REAP=1 to turn it off); and a fix for /model showing stale or empty state right after /login

The broader week

  • OpenAI previews the GPT-5.6 family - announced June 26, a limited preview of Sol (the flagship), Terra (balanced, which OpenAI says matches GPT-5.5 at roughly half the cost), and Luna (its fastest, cheapest tier). The headline additions: a new max reasoning effort that lets Sol think longer on hard problems, and an ultra mode that goes beyond a single agent, spinning up subagents to parallelize complex work. Sol also ships with OpenAI's most aggressive safety stack yet
  • OpenAI expands Daybreak with Patch the Planet - announced June 22, the cybersecurity program added an open-source-maintainer initiative (built with Trail of Bits, with cURL, Go, Python, and Sigstore among the early participants), shipped the full GPT-5.5-Cyber, and opened a partner program with Cisco, CrowdStrike, Okta, Palo Alto Networks, and Wiz. A security-and-compliance week on both sides of the fence
  • MCP's biggest revision yet lands as a release candidate - the 2026-07-28 spec is out as an RC ahead of its July 28 final: a stateless core that runs behind a plain load balancer (no more sticky sessions), server-rendered UIs via MCP Apps, a Tasks extension for long-running work, and authorization aligned with OAuth and OpenID Connect