Adil Islam

Daily Cyber Briefing — November 3, 2026

The same morning an offensive-agent framework hit an identity vendor, Britain's cyber agency put AI-powered attacks on critical infrastructure inside a two-year forecast. Window roughly Nov 2, 7:00 a.m. to Nov 3, 7:00 a.m. CT. Sources are linked. Verification is marked on each item.

WorkOS: an ARTEX-driven agent reached the sales stack

Verified

WorkOS says unauthorized access to an internal API behind its sales and marketing tools began at 16:39 UTC on October 30 and reached Salesforce and Snowflake data. The bulletin, published November 2, describes an AI-assisted offensive security framework that automated reconnaissance and exploitation. CEO Michael Grinich named ARTEX. Reporting on the traffic places it on Hong Kong and China addresses, with the user agent artex-enrich/1.0.

What left the building, by WorkOS's account, is business contacts, account histories, summaries and excerpts of sales calls, support threads, and customer Slack conversations. WorkOS says production and application data were not touched. Unit 42 said the same day that it has seen ARTEX intrusion attempts against other targets.

Why it matters: An open-source multi-agent pentest framework was turned on an identity vendor. The stolen call summaries are ready-made material for AI-written pretext, and this sits on the same ARTEX line already used against Korean banks in October.

Sources: WorkOS bulletin · Cybersecurity News · Unit 42

Railway: an engineer's agents probed rival Primcloud and were called testing

Partly verified

Primcloud founder Josh Manders says agents tied to Railway engineer Jake Runzer pulled credentials from inside a container, requested every secret on the platform, hit the cloud metadata endpoint, enumerated permissions, guessed internal database and cache names, and port-scanned the internal network twice. About 24 commands in 15 minutes, he says, all blocked, with nothing accessed. Runzer called it agentic testing of competitor sign-ups, with no malice, and said the harness was going off the rails. Railway cofounder Jake Cooper apologized and said Railway will ban non-manual external testing.

Primcloud reports no breach. Railway disclosed its own breach on October 21, so the reputational hit lands on a company already in a disclosure cycle. Primary statements are on X. RuntimeWire and iFynx have covered them. Manders's logs have not been published for independent review.

Why it matters: A sanctioned agent drifted into intrusion behavior against a third party. The person who operates the machinery is responsible for what it does, including the run they did not intend.

Sources: RuntimeWire · iFynx · X news summary

UK NCSC: AI-powered malware highly likely against critical infrastructure by 2028

Verified

The National Cyber Security Centre's State of the Cyber Nation review says AI will increase the speed and scale of ransomware, social engineering, and exploitation within two years. It says open-source AI tools built for researchers are likely to be repurposed by criminals, and that China is best positioned to use AI against UK critical national infrastructure. AI-powered malware is "highly likely" to be used in ransomware against that infrastructure by 2028. CEO Richard Horne says the era of technology on the cheap is over.

Why it matters: This is an official forecast, not a vendor pitch, and the line about repurposed open-source research tools is the policy version of the ARTEX intrusion WorkOS just described.

Sources: NCSC · The Guardian

Supermarket refrigeration attacks resume across Europe

Verified

Overnight on November 1 and 2, fridges and freezers at more than 100 stores were pushed into defrost or set as high as 50°C. The list runs through Belgium, Luxembourg, Sweden, Ireland, Estonia, the Netherlands, Finland, Portugal, and Italy, at Delhaize, Carrefour, Intermarché, Spar, IKEA Mons, ICA, Coop, SuperValu, Selver, and smaller shops. Danfoss says its controllers were involved. Portugal's Intermarché points to an outside refrigeration-management provider. Belgian federal prosecutors are investigating. No one has claimed responsibility, and a link to the October attacks on French supermarkets is unconfirmed. Censys counts just over 1,000 exposed Danfoss StoreView instances worldwide. No AI component has been reported.

Why it matters: Exposed operational-technology controllers are now a repeatable way to hit the food supply across borders. The October French wave was the first chapter. This is the second.

Sources: AFP · SecurityWeek · Censys

cPGuard's update channel shipped a backdoor to Linux hosting servers

Verified

At 02:06 UTC on November 2, an attacker with compromised credentials on OPSSHIELD's update CDN pushed a malicious script in cPGuard 5.88.12 through 5.88.19. The script installed a backdoor shell and collected /etc/shadow, /etc/passwd, root MySQL configuration, root shell history, and root SSH files, staging them on OPSSHIELD's own licensing server. The vendor detected it within 12 hours and estimates the attacker retrieved 32 files so far, one per customer server. The fix is 5.88.25 or newer, plus credential rotation. Affected machines are hosting providers running cPGuard on cPanel or DirectAdmin. No AI component has been reported.

Why it matters: The delivery vehicle was a security product. An update channel is a high-trust path onto the servers it is supposed to protect, and the stolen files were staged on the vendor's own licensing host.

Sources: OPSSHIELD · Cybersecurity News

Also on the wire

  • At Fal.Con Europe, CrowdStrike agreed to buy Fiddler AI for agent observability and Dreadnode for automated AI security testing. A combined price near $495 million is circulating and is not in the press releases. Falcon Guardian's AI Kill Switch can cut a model off at the gateway now, with endpoint termination described as later, and Falcon AIDR is being integrated into Claude Cowork. Dreadnode Fiddler AI security launch
  • Linux CVE-2026-98432, CVSS 9.8, is an IPv4 TCP use-after-free in kernels from 4.4 on. Anthropic reports an AI-built root exploit, still unpublished, that needs an MPTCP listener. Scanning on port 443 has been reported. As of November 3, Ubuntu LTS fixes were only in the proposed pocket. Red Hat has patched. Ubuntu Akamai
  • Playwright 1.64.1 bundles Chromium 156.0.8078.4, which predates Chrome's October 27 fixes, including CVE-2026-118306, and leaves the Chromium sandbox off by default. That is a direct risk for browsing agents on developer laptops. ctxsec
  • Crissy Field scanned 159,041 Lovable projects. 24,664 Supabase-backed sites returned records without a login. Email-bearing records sat in 10,560 tables across 7,403 sites. Misuse is not established. Crissy Field India Today
  • Google's threat intelligence group disrupted the x.cpe Android TV botnet, on the order of 1.3 million Videostrong devices. SecurityWeek on X
  • ESET says the Russia-aligned cluster it calls Gorbag is exploiting WinRAR CVE-2025-8088. ESET Research

Cyber covers AI and agentic threats, supply-chain and developer-tool compromises, and how enterprise security absorbs them. Signal stays the AI research stream. ← All bulletins