Security & Threat Intelligence
The Watch
Cisco's CVE-2026-76519 makes three exploited email gateway flaws in a year.
This one gives unauthenticated attackers root on an appliance that usually runs no EDR agent. If the unit was reachable before you patched, the only traces of an earlier break-in are its outbound traffic, any new admin accounts and changes to its config.
In Play
Edge and collaboration zero-days exploited
Risky.Biz and The Hacker News report active exploitation of Cisco Secure Email Gateway CVE-2026-76519, PAN-OS CVE-2026-0322 (CVSS 8.6) and Collabora/Nextcloud CVE-2026-110447. The Nextcloud flaw has been planting web shells since at least September. SANS @RISK saw exploit attempts on Atlassian CVE-2026-21589 one day after its Oct 5 patch. Patching does not remove an attacker who is already in, so each of these systems needs a compromise hunt.
AdFection turns ad rendering into spyware delivery
Citizen Lab found that Discover-Tech's Controlium platform let partners such as Candiru infect devices the moment an ad rendered, with no click. CyberScoop reports at least 45 Al Jazeera devices were infected, and the ads ran on 120,000+ websites and 70,000+ apps between March and June. Techpresso says NVIDIA and Arm were among the named targets, which puts technology-sector staff in scope. Your patch cycle, email gateway and awareness training never see this channel.
Agent platform defaults leak credentials and spread payloads
Zenity Labs showed that one chat prompt to an AWS Bedrock AgentCore agent could pull IMDS credentials that compromise every AgentCore agent in the same account and region. tl;dr sec also covered research showing that hidden white text makes Copilot for Word copy its payload into downstream documents; two Microsoft mitigations over 144 days did not stop it. CSO reports that Claude Code's HTML preview leaked developers' cloud credentials. In all three cases the vendor's default settings were the weakness.
Developer credentials and AI-recommended repos are the cheap way in
GitGuardian, as reported by The Hacker News, says the GhostAction crew compromised about 1,900 GitHub accounts on Oct 8-9 and targeted 30,000+ repositories. Risky.Biz reports that 88 fake GitHub repos spread the MicrosoftSystem64 RAT. Island's research shows ChatGPT and Google recommended one of those repos based on forged stars and forks. The Aoki infostealer now also targets AI coding assistants, so the developer workstation is where stolen credentials start.
AI labs are moving into your security vendor chain
The Information reports that OpenAI is in early talks to buy Trail of Bits, a security firm it already works with. It also reports that Cursor staff investigated whether confidential data reached SpaceX for AI training; the outcome of that investigation has not been reported. If either story holds, an AI vendor would sit inside your assessment or code pipeline, so independence, no-training and change-of-control terms now carry real weight.