Unit 42 reports PAN-OS exploitation reaching virtualization infrastructure

Independent

Unit 42's October 19 update reports ransomware operators exploiting CVE-2026-0257. In one case, attackers moved from a compromised VPN gateway to vCenter, disabled an ESXi security control and performed offline credential dumping against a domain controller's virtual disk.

Why it matters: For exposed GlobalProtect operators, the reported path extends from authentication bypass to credential theft outside the guest operating system. Unit 42 says this bypasses in-guest endpoint protection, making virtualization infrastructure relevant to compromise assessment.

Decision: If investigating successful VPN connections linked to this activity, extend the assessment to vCenter and hypervisors rather than relying solely on guest endpoint telemetry.

What remains uncertain: These are vendor investigation findings, not independent confirmation; victim counts are undisclosed and the virtualization technique is illustrated by one case. The brief retains an earlier no-lateral-movement observation, superseded by its explicit October update. Advisory exposure requires a GlobalProtect portal or gateway, enabled authentication override cookies and a specific certificate configuration. Panorama and Cloud NGFW are excluded.

Read: Reported article: Palo Alto's firewall OS has been under attack from multiple ransomware groups since June; Primary source: CVE-2026-0257 PAN-OS: GlobalProtect Authentication Bypass Vulnerabilities; Primary source: Threat Brief: Active Exploitation of PAN-OS CVE-2026-0257 (Updated October 19)

Researcher reports one operator directing AI agents against more than 1,200 sites

Independent

NorForge researcher Snorre Fagerland reports that one operator used AI agents to compromise websites belonging to more than 100 companies, estimating US$5,000 in token spending. Reuters independently reviewed the operator's activity to establish victim identities; several businesses acknowledged incidents.

Why it matters: Fagerland describes sustained autonomous investigation: successful intrusions typically took around 23 hours, and one agent workflow found an associated control panel after failing against the initial application. For public-facing website operators, this suggests a single adversary can sustain probing across many targets with limited direct labour.

Decision: Include associated public-facing control panels when defining the scope of an external application security review; the reported intrusion crossed that boundary.

What remains uncertain: Autonomy, costs and campaign scale largely rest on Fagerland's observations; his underlying report is not supplied. Attempts against 1,200 sites are not 1,200 breaches. Reuters could not establish the extent of data extraction. Deere reported no indication of sensitive-information access; SharkNinja described a contained incident on a standalone support site. Cummins was investigating a potential attack on a vendor-operated website.

Read: Reported article: How one hacker used AI agents and a few thousand dollars to break into scores of sites - iTnews

No linked primary source could be read; claims rely on the reporting above.

curl plans public handling of low- and medium-severity vulnerability reports

Independent

From October 26, curl plans to route low- and medium-severity reports to public GitHub issues, reserving private handling for high and critical flaws. Maintainer Daniel Stenberg attributes the change to overwhelming report volumes; researchers uncertain about severity can still submit through HackerOne.

Why it matters: Stenberg expects public handling to broaden participation in assessment, fixes and advisories, while acknowledging that it exposes flaws before fixes arrive. For organizations embedding curl, this changes where vulnerability information appears. He says over 90 percent of submissions over the last few years were quickly assessed below high severity, suggesting most submissions could follow the public route.

Decision: For teams monitoring curl vulnerabilities, include public GitHub reports in assessment rather than waiting exclusively for release advisories, checking whether reported option combinations apply to deployed uses.

What remains uncertain: This is an announced policy, not a measured capacity improvement. Stenberg acknowledges that work per report will not decrease. His original statement is not supplied. The article acknowledges useful AI-assisted reports but does not quantify how many submissions describe exploitable deployment risks.

Read: Reported article: Swamped curl maintainers take low-level flaws public to survive vulnerability 'avalanche'

No linked primary source could be read; claims rely on the reporting above.