Practical defense
against AI-era attacks

Attackers are using AI to move faster. We break down real incidents, show how to defend with AI agents, and lay out the data-handling procedures that stop a breach from spreading.

Latest incidents and guides

Identity log showing fresh Azure tokens issued, beside the two compromised agent-framework versions

Compromised Microsoft agent-framework releases on PyPI ran a credential stealer on every Python start, aimed at Azure and CI secrets

Two malicious agent-framework versions were up for under an hour on Nov. 17. Beyond stored secrets, they requested fresh Azure tokens, so identity logs need review.

News · 3 min read

Ten squares, three in blue marked major: nginx 1.31.9 fixes 10 CVEs, and a reload does not apply the fix

nginx 1.31.9 fixes 10 flaws and needs a restart; researchers ran code through the HTTP/2 bug in lab tests

Exposure depends on configuration and on embedded copies you may not track. A reload does not apply the fix; community ingress-nginx gets none.

News · 3 min read

GTIG case study page highlighting passkeys that persisted across resets, with a case 1, case 2 and VirusTotal link strip

Google: Russian spy group COLD RELIC posed as a breached organization's employee to break into a second US organization

GTIG says COLD RELIC impersonated an employee of one breached US organization to reach another. A password reset did not evict it in at least one case.

News · 3 min read

Java runtime exposure check: JDK train, AVX-512 VBMI x86-64 CPU and attacker Base64 must all hold; interim flag shown

Java4Shell: a Base64 flaw in OpenJDK lets attackers run code on newer Intel and AMD servers, researchers say

Three conditions decide whether a Java system is exposed. The hard part is finding runtimes bundled inside vendor products you cannot patch yourself.

News · 3 min read

Run view of five steps from vendor patch to webshell scripts, beside our card: start the patch clock at vendor release

Google says one attacker framework diffs web-app patches and runs the exploits without a human

Google's threat intelligence group describes a framework on one exposed server that turns patches into exploits. We would start the emergency patch clock at the vendor's release, not at a KEV listing.

News · 3 min read

Rack-mounted security gateway with labels: PHP2Shell entry, root flaw first fixed Nov. 12, hotfix closes entry only

Check Point says attackers chained PHP2Shell with a root flaw it first fixed on Nov. 12

Attackers entered unpatched Check Point Mobile Access portals through PHP2Shell, then took root. The October hotfix closes only the entry; the November Jumbo takes close both where Check Point has released them.

News · 3 min read

Browse by topic

Defense first

We explain how an attack works so you can stop it. We do not publish working exploits.

Facts checked against their sources

Claims are checked against their sources before publication. Versions, prices and other values that change are re-checked afterward.

Responsible disclosure

Found a flaw in something we cover? Write to the editors at contact@cybersecurity.guru.