CISA flags three Linux kernel flaws as actively exploited; critical pre-auth RCE hits Orkes Conductor

CISA adds three exploited Linux kernel flaws, including a 9.8 in the TLS receive path

CISA on Friday September 19, 2026 added three Linux kernel security flaws to its Known Exploited Vulnerabilities catalog, citing evidence of active exploitation. The list includes CVE-2025-39682, scored 9.8, described as an improper check for unusual or exceptional conditions in the TLS receive path. Adding items to the KEV catalog triggers federal remediation timelines and forces operators of regulated infrastructure to treat the flaws as priority work, not backlog. Practitioners running Linux on internet-facing services should treat TLS-path defects as immediate-fire items regardless of distro version, since KEV listing is independent of any one vendor's advisory.

Critical pre-auth RCE in Orkes Conductor is already being exploited

A critical unauthenticated remote code execution vulnerability in the Orkes Conductor workflow platform is being exploited in the wild, according to Fortinet. The flaw is tracked as CVE-2026-58138, with a CVSS v3.1 score of 9.8 and a CVSS v4 score of 9.3. Affected versions are Orkes Conductor 3.21.21 before 3.30.2; operators must be on 3.30.2 or later to be patched. Workflow engines are high-value targets because they glue together authentication, payments and data movement, so a pre-auth RCE that is already under exploitation is the kind of bug that turns into a weekend incident for any shop that runs Conductor on its perimeter.

SolarWinds patches hard-coded key enabling unauthenticated RCE in Access Rights Manager

SolarWinds released security updates for Access Rights Manager to address a high-severity flaw that could lead to unauthenticated remote code execution. The issue, CVE-2026-28326, is rated 8.8 on CVSS and affects all versions of Access Rights Manager 2026.2 and prior. Because the underlying defect is a hard-coded key rather than a memory-safety bug, the patch is not optional: any deployment still on 2026.2 must move to the fixed build before the next credential-rotation cycle. Operators should pair the upgrade with a search for stale service accounts, since hard-coded-key bugs typically compromise everything the affected component can authenticate to.

Supply-chain and credential reuse spread from May's TanStack npm attack

CrowdSec disclosed on September 18, 2026 that an attacker copied about 170 of its private GitHub repositories on May 22 using the account of a recently departed employee. The employee's GitHub access had not been revoked, and his laptop was compromised in the May TanStack npm supply-chain attack, in which malicious TanStack packages stole credentials from developer machines. The incident illustrates a recurring failure mode: offboarding procedures that leave production SCM credentials valid for months after an employee leaves, and developer endpoints whose secrets were siphoned by an upstream package compromise. Treat both as separate checklist items in the same review.

Crypto theft, ransomware leak-site breach, and AI-driven exploitation widen the threat surface

7 million in stolen cryptocurrency to North Korea. 6 million user records, and the ShinyHunters extortion group breached the Clop ransomware operation's data-leak site, defacing it and reportedly stealing private keys for its onion service. Together the items sketch a week in which state-aligned theft, image-host breaches and ransomware-on-ransomware attacks landed within 48 hours, and every one of them hinges on weak secrets, unpatched servers, or unrotated keys rather than novel cryptography.

AI agents as a new attack surface: extensions, domain mix-ups, and the disclosure-to-exploit gap

Forever Security researcher Gal Weizman demonstrated "BragJack," a proof-of-concept attack that hijacks the AI assistants in Chrome, Edge, Opera Neon, Perplexity Comet, and Claude in Chrome through a single malicious browser extension using a Prompt Forcing technique, earning over $20,000 in bounties and two CVEs. In a separate evaluation run in May 2026, Google's Gemini model accessed real company systems during a cybersecurity test by Israeli company Irregular after a domain mix-up. Security programs still validating CVE risk on weekly or quarterly cycles are now racing AI tooling that compresses the time between disclosure and working exploitation.

The lesson for practitioners is to instrument browser extensions in the same tier as local binaries, and to require identity verification before any agent acts on internal systems.

What to check next

2. Then audit GitHub and other SCM accounts belonging to anyone who has left the company since May 2026, and rotate any secrets that lived on a machine that ever ran npm install during the TanStack incident window. Finally, review which browser extensions your developers and operators have installed in Chrome, Edge, Opera Neon, Perplexity Comet, and Claude in Chrome, since the BragJack PoC shows one extension can steer multiple agents at once.

No deadline for the AI evaluation work or pending Cloudflare Pingora memory optimization was named in the evidence, so treat those as ongoing engineering items rather than dated commitments.