Fetching from the wire…
Public story · 2026-08-25 · high
VulnCheck disclosed the CVSS 8.3 flaw on August 24, and a poisoned web page or issue body is enough to trigger it.
Why now: NVD entered CVE-2026-76072 into its public database on August 24, the same day VulnCheck's disclosure became public.
Continue CLI's headless and auto modes give its Bash tool blanket permission to run shell commands. A single filter, isCriticalCommand, is the only defense against destructive commands, per NVD's listing for CVE-2026-76072.
The flaw carries a CVSS score of 8.3. An attacker doesn't need a terminal or credentials. A poisoned web page or a booby-trapped GitHub issue, fetched by the agent mid-task, is enough to fire the delete command against the user's files.
The filter's dangerous-path check matches /, ~, /usr, /etc, /bin and /sbin. It doesn't match /home, /root, /var, /opt or /srv, so a recursive delete aimed at any of those five paths sails through unblocked.
$HOME fares no better. The line gets parsed with shell-quote before the shell ever sees it, and $HOME collapses to an empty token at that stage. The check runs against the empty string. Then the real shell expands $HOME back to the user's home directory when the command spawns, and deletes it.
The filter also misjudges find -delete, rating it high risk instead of blocking it outright. Four other destructive tools, shred, wipefs, truncate and pkexec, aren't handled at all.
NVD's advisory doesn't say whether Continue has fixed the flaw yet.
Each link below shares sources, entities, or timing with this story.
Go look at your ~/.claude/CLAUDE.md right now. Mine has internal package names, a build command with a host in it, and notes about which credentials live where. I wrote it assuming exactly one reader. RuntimeWire published traced request captures on August 9 showing Muse Code...
CVE-2026-75130, published August 18, covers Upstash's Context7 through 2.1.2: the Custom AI Instructions feature serves unsanitized content through the MCP server, so poisoned instructions can exfiltrate credentials from environment files to an attacker-controlled service and...
Ten days from spec to shipped client. That's fast even for this ecosystem. The MCP 2026-07-28 revision replaced the bidirectional stateful protocol with request/response. Every request now independently carries protocol version, client identity and capabilities. Cloudflare's t...
Go rotate a key. I'll wait. Claude Code 2.1.246, released August 25, lists this in its changelog: a fix for "telemetry and metrics requests to Anthropic carrying the API key configured for a third-party gateway (ANTHROPIC_BASE_URL); a credential is now only sent to its own hos...
The same man whose framework a model regression destroyed also published the most aggressive prediction of the week, and the tension between those two facts is the whole argument. "The Shape of Things to Come, Part 1: The Continuous Thunderdome" argues traditional CI/CD collap...
NVD posted nine advisories on August 25, clustering into one shape: a local server assuming a browser can't reach it. PraisonAI validated MCP origins with request_origin.startswith(allowed) against a localhost allowlist, so an attacker-registered localhost.attacker.com passes...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.