BroadWorks has emerged as the immediate patching priority. KEV status tells us that both CVE-2026-20320 and CVE-2026-111793 have evidence of active exploitation, but it does not establish campaign scale, universal exposure, or a complete path to privileged execution. The European telecom report adds real-world impact by confirming access to sensitive information, while still leaving command execution, persistence, and broader operator compromise unproven. The distinction between the two flaws matters operationally: BroadWorks presents an unauthenticated, remotely reachable blind XXE with no workaround, whereas the Openwave issue requires an authenticated user and permits bypass of file-type restrictions in the transcoding workflow. That supports Alex’s call to put BroadWorks first, without overstating what either vulnerability alone gives an attacker.
The Railway incident gives us a different but equally important trust-boundary lesson. A tenant build crossed into partial host access, exposed host credentials, and then benefited from erroneous OIDC trust to reach short-lived cloud credentials. The confirmed scope is 353 accessed build histories and roughly $740,000 in unauthorized spending, principally on AI services. The possible decryption exposure affecting up to 1.85% of workspaces is an upper bound, not a count of proven secret theft. Railway owns the failed isolation, host-credential, and federation controls; customers still need to rotate credentials entrusted to the platform and investigate their downstream use.
For Arcus Perps, the likely broken invariant is that transfers among subaccounts must never increase aggregate account equity or create withdrawable collateral. The reported 2,234,391 USDG withdrawal establishes protocol loss, but not automatically attacker profit or laundering; those require a complete on-chain reconciliation and evidence of subsequent obfuscation or cash-out. Treasury reimbursement may protect users from an immediate shortfall, but it does not repair the accounting defect. And in the 911 hoax, the bomb threat is clearly established as false, while synthetic-audio attribution remains only moderate-confidence because no recording, forensic analysis, provenance data, or carrier evidence is public. Verification should therefore run alongside emergency response, never delay it.
James now needs to turn these distinctions into a defensible operating sequence: immediate BroadWorks containment and patching, evidence-led compromise assessment, and controls that do not confuse active exploitation with proven post-exploitation depth.