"Black October": AI-driven attack framework breaches ConnectWise and ripples into a wider vulnerability-disclosure wave
2026-10-22
Microsoft attributed a supply-chain breach of IT platform ConnectWise to a China-aligned actor ('Storm-0940') using an open agentic pentest framework, reportedly compromising more than 2,250 Microsoft 365 tenants at roughly 2,900 sign-in attempts per minute and reaching privileged access in under three hours in some cases, per TechCrunch. CrowdStrike warned, per CIO Dive, that agentic AI is making low-skilled hackers more dangerous without detailed instructions, and CIO Dive separately reported JPMorgan Chase and Goldman Sachs executives emphasizing AI security balance and tech executives citing overreaching autonomous agents and unclear accountability as top concerns. A Latent Space digest of social-media reporting adds, without independent verification in hand, that in the same roughly-48-hour window a critical unauthenticated PHP remote-code-execution bug ('PHP2Shell') and a CVSS-10 GitLab flaw were disclosed, that the PHP bug was reportedly found by OpenAI models in August and embargoed until a Wednesday fix, and that monthly CVE disclosure volume has risen from about 4,000 to about 15,000 since April.
Significance 4: The ConnectWise/Storm-0940 attribution is corroborated by Microsoft via TechCrunch and backed by a named-firm (CrowdStrike) warning via CIO Dive, which is a real, verifiable, large-scale agentic attack. The broader PHP/GitLab/CVE-volume cluster rests mainly on a single curator's relay of unlinked social-media reports, so that portion is held as supporting context rather than independently verified fact.
Implications · machine-drafted, not owner judgment
This is concrete evidence that agentic AI is compressing both sides of the vulnerability lifecycle — discovery and exploitation — faster than enterprise patch cycles can keep up, which raises the floor for what counts as basic security hygiene and strengthens the already-fundable case for agent-security startups (AIUC, Baselayer, Reco, Nvidia's platform, Cohere's AISLE acquisition, and others already tracked). For the owner, this is a textbook operational-lane 'process friction' breakpoint: the technology (AI-driven discovery and attack) is outrunning the organizational process (patching, disclosure, agent-identity tracking) around it, and the labs whose models are finding these bugs (OpenAI; disputedly Anthropic) have no clearly disclosed coordinated-disclosure framework yet.
- Whether OpenAI or Anthropic publishes a formal coordinated-disclosure policy for AI-discovered vulnerabilities
- Independent confirmation of the 4,000-to-15,000 monthly CVE volume claim from a named CVE-tracking source
- Further confirmed attributions of AI-agentic attack frameworks beyond Storm-0940/ConnectWise
Sources
- press Chinese hackers spied on thousands of companies after breaching IT giant ConnectWise retrieved 2026-10-23
- press Agentic AI helps hackers level up — even without detailed instructions retrieved 2026-10-23
- press JPMorgan Chase, Goldman Sachs underscore security in AI plans retrieved 2026-10-23
- press What's keeping 5 tech executives up at night in the AI era retrieved 2026-10-23
- curator [AINews] Black October retrieved 2026-10-23