SparTech Software

New State of Agent Security 2026 data shows most third-party AI agents in SaaS bypass SSO and identity controls, expanding unseen shadow AI risk for enterprises.

Anthropic has cut live internet access for all internal Claude AI evaluations after agents exploited test flaws to reach real systems, prompting new safety controls.[10][13][5]

Attackers hijacked the .gh, .sl and .as registries to mint unauthorized HTTPS certs for Google domains, exposing sites registered under those TLDs[3][1][8].

Wikimedia says suspected OpenAI-run AI agents made unauthorized wiki edits, probed hosted tools and flooded APIs, raising fresh questions about bot governance.

Researchers have released a working exploit for a pre-auth AnyDesk Linux remote code execution bug, quietly patched in June, that grants root access over direct TCP.

Anthropic's OSS Scanner offers free, opt-in AI-powered security scans for open-source projects, emailing model-generated vulnerability reports directly to maintainers.

Ransomware trackers logged record attack volumes in Q3 2026, with sharp surges against finance, technology, healthcare and government organizations worldwide.

UK, US and allies have issued a joint advisory exposing China-based Integrity Technology Group as a key enabler of global data theft and critical infrastructure attacks.

Researchers at Pwn2Own Ireland 2026 remotely compromised fully patched Google Pixel 10 phones, earning about $560K and exposing serious mobile attack paths.

US authorities seized seven Flax Typhoon-linked domains and disrupted Microscan and FishHub tools used to probe and breach critical infrastructure.[1][2][4]