Someone forked a Tor-based agent skill, added one zip file, and the fork led six layers down to an infostealer, with its fallback C2 address hidden in a Polygon smart contract.
Our Delta team unpacked it and wrote up what to hunt.
spectrum.security/blog/fake-…
This phishing kit has no credential-capture code. It streams a real Google sign-in from the attacker's browser, so your password and MFA approval land in their session.
Our Delta team took it apart and wrote the detections.
spectrum.security/blog/brows…
Thanks to those of you who let us interview you at @BlueTeamCon!
We had a blast sponsoring - we'll see you next year.
Approving an MCP server tells you nothing about what it does tomorrow: the manifest can stay byte-identical while only the code underneath changes.
Six MCP attacks, built and run by our Delta team.
spectrum.security/blog/mcp-m…
By 7pm on day one of Fal.Con you'll have walked nine miles of casino carpet and heard "agentic" more times than is medically advisable.
Tee Time is the cure. 7-9pm Sept 1, Topgolf Las Vegas, with @Tailscale.
spectrum.security/events/fal…
Meet Delta - Spectrum's skunkworks research team, making AI tooling so that the defenders have a chance.
spectrum.security/blog/intro…
Somewhere in Vegas on Sept 1 there's a blackjack table waiting to take your money. Here's an alternative.
Tee Time at Topgolf, 7-9pm during Fal.Con, with @Tailscale. Golf, drinks, food, and you don't have to be good at golf.
Come shank a few with us!
spectrum.security/events/fal…
QRadar EDR and XDR hit end-of-life Aug 31, so a lot of teams are migrating detection content this month. A converter can translate rule syntax. It can't tell you the translated rule still matches anything. Imported is not the same as firing.
IBM's new breach report: 1 in 4 malicious breaches are now AI-enabled (up 56%), averaging $6M. Defensive AI cuts costs by ~$2M, but only on top of detections that actually fire. Attackers adopted AI faster than detection content did.
An all-green MITRE ATT&CK heatmap is the security version of a test suite that passes because someone commented out the tests. Green means a rule exists, not that it fires. And nobody's ever been breached by a technique that showed up red.
A GitLab RCE exploit went public last week, six weeks after GitLab shipped the fix as a routine bug fix (no CVE, no advisory). If you triage off CVE feeds, you never saw it. The only safety net left is a detection that actually fires on the attack.
Happy Friday to everyone except the detection rules that quietly stopped firing the day Microsoft renamed AADSignInEventsBeta to EntraIdSignInEvents.
One of your detections probably stopped firing weeks ago, and nothing in your stack told you.
Picus found that half of all detection failures come down to log collection quietly shifting underneath the rules, blinding them without a sound.
thehackernews.com/2025/08/wh…
