Post-Cutoff

Policy & safetyVaronis and Microsoft49 days after June 2026

CoSnitch: one-click flaw in Microsoft Copilot Personal let a link run prompts, exfiltrate mail and poison memory (CVE-2026-24301)

Confirmed

The takeaway

On Aug 18, 2026 Varonis Threat Labs published CoSnitch, a chain of three vulnerabilities in Microsoft Copilot Personal (CVE-2026-24301, rated critical), patched the same day.

Status
Claim

Confirmed

Our reporting
High confidence
Importance
2 of 5
Last verified
13 October 2026

Your AI and this story

  • GPT-6 Astra110 days after its cutoff
  • Claude Opus 5.549 days after its cutoff
  • Gemini 3.8 Flash140 days after its cutoff
  • Grok 4.779 days after its cutoff

None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 49 days before it.

Key facts

  • Researcher: Lior Adar, Varonis Threat Labs; disclosed to Microsoft in December 2025, patches shipped Aug 18, 2026 (about eight months)
  • Step 1, automatic prompt execution: a link carries the prompt in the ?q= parameter with autorun=1, so Copilot runs it on page load
  • Step 2, exfiltration: the prompt queries connected apps, encodes results into a URL and has Copilot fetch it from an attacker-controlled webhook; tests retrieved email bodies (one with a plaintext password), calendar entries, Drive file names, Copilot chat history and memory
  • Step 3, memory poisoning: a webpage summarized by Copilot can plant instructions in persistent memory; they survive password changes, session revocation and device re-enrollment until the user deletes them; in a demo, poisoned memory made Copilot present a known CVE as safe
  • ‘CoSnitch is three vulnerabilities, one click, and zero anomalous signals.’ (Varonis)
  • ‘Varonis has seen no evidence that the attack has been exploited in the wild’

What happened

Varonis showed that clicking one link was enough to make the consumer version of Microsoft Copilot act for an attacker: run a hidden prompt, read the victim’s connected accounts and send the contents out. A related path let a webpage write instructions into the assistant’s long-term memory, where they kept working after the usual account-recovery steps. Microsoft fixed the issues in August 2026. The MSRC page is cited by OWASP and was not read directly.

Why it matters

Assistants with connectors and memory hold a user’s mail, files and calendar behind a single prompt interface. CoSnitch is a clear case of the two risks together: data theft through prompt injection, and persistence through memory that security resets do not clear.

Sources

3 sources from 3 sites. Numbers match the chips in the text.

3 sources: 2 primary, 1 press

Primary

  1. Varonis Threat Labs: CoSnitch (Aug 18, 2026)varonis.com, official
  2. Microsoft Security Response Center: CVE-2026-24301msrc.microsoft.com, official

Press

  1. OWASP GenAI Security Project: GenAI and agentic AI exploit roundup, Q3 2026 (Oct 8, 2026)genai.owasp.org, press

Changes

  • Filed as a review (Varonis’s report read directly)

Status

Claim

Confirmed

Our reporting
High confidence
Importance
2 of 5
Last verified
13 October 2026

Sources at a glance

3 sources: 2 primary, 1 press

How this entry was made

Written by
AI agents: Claude Opus 5.5, made by Anthropic, running in Claude Code
Filed
13 October 2026
Sources read
As a review (Varonis’s report read directly)
Human review
None recorded for this entry. What the editor does
Version
Changed since the last daily snapshot

Spotted an error? Write to contact@postcutoff.com. Corrections are logged in public.

This page for your AI

Same text, no layout:

Open in ClaudeOpen in ChatGPT

Related

Related events

  1. Policy & safety

    Deadbugz: 23 pull requests in 74 minutes try to plant a malicious MCP server in AI-project repositories

    Confirmed