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

Research page quoting Dragos: a privacy flag only binds the first hop; prompts go via router and are copied at the provider

OpenRouter drops three inference providers and pulls 125 endpoints from Zero Data Retention after Dragos canaries showed prompts were copied

OpenRouter says it found no breach of its own systems. Providers behind it have published windows in which prompts may have been exposed. Rotate secrets that went through them.

News · 4 min read

Library announcement page highlighting that GPGME is affected as a shared library, a DLL or a static link

GnuPG calls a GPGME double free critical, and a package update will not reach every copy

Static, bundled and image copies of the library need rebuilds, not just a package update. Here is what a scanner will miss and which copies to fix first.

News · 3 min read

Medical imaging AI server with authentication off; agent sessions: file write in about 11 minutes, root later

Amazon says Cairn AI agents found an unreported flaw in an open-source server and got root

Amazon says agents given target IPs and variations of the goal "obtain shell access" found a file-write flaw in a medical imaging AI server. No fixed release exists, so exposure and authentication are the controls.

News · 3 min read

nginx.conf with fastcgi_keep_conn on highlighted, beside the order: check reuse, turn off, restart PHP-FPM, upgrade

PHP's November fixes close gaps in the PHP2Shell patch; check FastCGI connection reuse first

Two of PHP's 17 advisories from Nov. 19 describe gaps in the October PHP2Shell fix, and a third bug was made reachable by the release that shipped it. Whether your web server reuses FastCGI connections decides which hosts to upgrade first.

News · 3 min read

Rack servers with labels: exploited before Nov. 19 advisory, forged SAML bypasses MFA, guest disks read for NTDS.dit

Google says a suspected China-nexus group exploited vCenter from early September, before Broadcom's Nov. 19 advisory

GTIG links CVE-2026-123890 to a chain of forged SAML tokens, ESXi logins and domain controller disk reads. Patch where Broadcom has a fix, then hunt back to September and limit who reaches vCenter; 7.0 has no public fix.

News · 3 min read

Research post page, highlighted: a database's SQL privilege model collapses into commands as the database's OS account

Minimal PostgreSQL and MySQL logins reached the server's OS in depthfirst's tests: database grants are not an OS boundary

depthfirst turned single memory bugs into commands as the database's OS account from minimal PostgreSQL and MySQL logins, and says its Oracle chain needs no login. Patch, revoke the default temporary-table rights, and strip the host of what it can read.

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.