Bell Canada Says an 'Agentic' Attacker Took Enterprise Customer Data. The European Commission Calls Its Attack 'AI-Generated.' Neither Says What That Means. Here Is What to Change Anyway.

If you are a Bell enterprise customer, assume the people who handle your Bell account will get convincing phishing in the coming weeks. Bell says the stolen data includes business contact details, ticket titles and some IP addresses of network elements. That is enough to write an email that names your account rep, quotes your own ticket and refers to your circuit. Tell those people now that Bell will not ask them for credentials or payment changes by email, and confirm any request through a number you already have. Then rotate any password or API key you have ever pasted into a Bell ticket or portal, even though Bell says no credentials were taken.
Two large organizations confirmed this week that an attacker using AI tooling got into their systems. Bell Canada called its attacker "agentic." The European Commission's spokesperson called its incident "an AI-generated cybercampaign." Neither has explained the word. Both statements are a company's description, not a technical report, and we treat them that way below.
Am I affected?
- Bell enterprise customers: look for a notice from Bell's Information Security Team sent this week. BleepingComputer saw one through a reader. If you got it, you are in the population. If you did not, ask your Bell account team directly rather than assuming you are out.
- Your own systems, the question this week raises: do you have a pre-production or test copy of a customer-facing system that holds real customer records and can be reached from the internet? Bell's notice says the attacker reached "a pre-production environment of Bell's Maximo system." Search your asset inventory for hosts with names like dev, test, uat, preprod or staging that answer on the internet, then check what data each one was loaded with.
- Can you see an agent in your logs? Pick your busiest customer or partner portal and ask whether you could tell, from its logs, one session that walked through hundreds of records or endpoints in a few minutes. If the answer is no, that is the first thing to fix (see below).
Bell Canada: what Bell said
Bell's statement to BleepingComputer and to The Globe and Mail (spokesperson Ellen Murphy, by email):
- "Bell is investigating a cybersecurity incident that occurred on November 17."
- "an agentic threat actor accessed some of our systems, including a system used for ordering and provisioning circuits." The Globe's copy says "business circuits."
- "Upon learning of the breach, we immediately contained the incident and secured our systems."
- "There has been no impact to customer services."
- "We currently do not know the attribution."
The customer notice, as BleepingComputer quotes it, says "an unauthorized party" accessed "a pre-production environment of Bell's Maximo system," which Bell uses to order and provision network connections. The notice itself does not use the word "agentic." That word appears only in Bell's statement to the press.
Data Bell lists: business contact information (names, job titles, corporate email addresses, phone numbers, business addresses), information in ticket titles, some IP addresses tied to network elements or systems, and "limited contextual information."
Data Bell says was not taken: credentials, highly sensitive personal information, PDFs, file attachments and other supplementary Maximo information.
What Bell is doing: notifying affected enterprise customers, working with the Canadian Centre for Cyber Security, and sharing indicators of compromise and threat-hunting information with enterprise customers and the Cyber Centre. Bell has not published those indicators. It warned customers about unsolicited requests from people impersonating Bell.
What is not known: how the attacker got in (no entry point is named), how many customers are affected, how much data was taken, who did it, and what made the attacker "agentic." The Globe and Mail says Bell gave no details on that last point. BleepingComputer writes that the word "suggests" AI-powered tools were used. That is the outlet's reading, not Bell's finding.
One source. Every report we found (BleepingComputer, The Globe and Mail, SC World and the aggregators that copied them) traces to Bell's emailed statement and the customer notice a reader passed to BleepingComputer. We found no public statement on Bell's own site. SC World's page returned HTTP 403 to us.
What Bell enterprise customers should do
- Brief the people whose names were in the data. Account contacts, network engineers and accounts payable are the likely phishing targets. The stolen fields (title, email, phone, address, ticket titles) are exactly what a pretext needs.
- Rotate what lives in tickets. Bell says credentials were not taken, but ticket titles were, and engineers sometimes put more in a ticket than they should. Rotate any password, pre-shared key, SNMP community or API key that was ever shared with Bell in a ticket or portal message, and the logins your staff use for Bell's enterprise portals. Turn on MFA for those portal accounts wherever Bell offers it.
- Treat the exposed IP addresses as recon. If Bell's provisioning records held management addresses of your routers or circuits, an attacker now has a target list. Check that those management interfaces are not reachable from the internet and that their logins are not defaults.
- Ask Bell for the indicators. Bell says it is sharing them with enterprise customers. Get them, search your firewall and proxy logs back to at least November 1, and ask Bell what time window to search.
European Commission: a new incident, not the March one
This is not the incident we covered in April. That one was the March breach of the cloud infrastructure behind the Europa.eu websites, which CERT-EU attributed to TeamPCP after the Trivy scanner compromise (our April post). The Record counts the new incident as the Commission's third disclosed this year, after January's Ivanti breaches and March's cloud breach, and says officials have not linked them.
What the Commission said. We found no press release. The statements come from spokesperson Thomas Regnier at the Commission's November 19 press briefing, as Reuters quotes him, and from a spokesperson's comments to Recorded Future News and to Politico:
- "I can confirm that the Commission has recently taken necessary steps in response to a detected cybercampaign."
- "As a protective and preventive measure, we temporarily took down several applications," and "most of them have since been restored."
- "This was an AI-generated cybercampaign."
- "The investigation is still ongoing."
- To Politico: "Individual systems were targeted and, in line with our standard practice, we have informed those concerned."
Regnier said the first signs were detected on October 22, that there was no evidence of a breach affecting Commission staff as a whole, that he knew of no ransom demand, and he declined to name a suspect. Reuters notes that "AI tools" in its own headline is its wording; "AI-generated cybercampaign" is the only AI phrase from the Commission.
What came from Politico's sources, not the Commission. Politico reported first, in its Brussels Playbook newsletter. Its sources say the disruption lasted more than ten days and hit Sysper (payslips, timesheets and leave), MyPMO (medical reimbursements) and translation services, that large amounts of staff personal data may have been exposed, and that CERT-EU identified the attackers' use of ARTEX, a public AI penetration-testing framework. The Register reports Sysper went down again on Wednesday, November 18. The Commission has not itself named ARTEX. Nobody has said how the attackers got in, or whether ARTEX was used to get in or only to look for weaknesses.
What CERT-EU published. CERT-EU's November 5 note, "From Penetration Testing to Cyber Threat: Beware of ARTEX," calls ARTEX "an open-source, LLM-driven autonomous penetration testing platform" and says it "has observed similar instances of AI-driven activities against its constituency," which "highly likely involved the use of ARTEX." It does not mention the Commission incident, and it publishes no indicators.
We have covered ARTEX several times, most recently the FBI's November 13 FLASH and its 22 addresses. That post has the detection and hardening list from the FBI; we will not repeat it here.
What changes when the attacker is an agent
Strip the adjective and both incidents are ordinary: something internet-reachable was entered, and records left. The agent changes the timing and the volume, and both are measurable.
Speed. The FBI's ARTEX FLASH says some intrusions went "from initial targeting to privileged access in under three hours." A patch or exposure window you measure in days is now longer than the whole attack. For internet-facing systems, the useful target is the same day: when a system shows up exposed, or a fix for an exploited bug is released, the clock is hours, not the next change window.
Volume. An agent tries more things than a person, cheaply. NorForge's look inside Cairn operators' panels estimated about $5 in model fees per attack task (our October post). Systems nobody watches, like a pre-production copy of a provisioning system, are the ones it will find. Bell's notice names exactly that kind of system.
The agent as a signal. The same speed that hurts you is visible in logs, if you log the right fields:
- Rate and breadth per session. The FBI asks defenders to rate-limit important APIs and alert on sequential lookups or unusually many records per session on partner and employee portals. A person reads a few tickets; an agent enumerates them.
- Tool patterns. CERT-EU tells defenders to hunt for "automated scanning patterns, specific tool signatures." In practice: one source trying many paths, parameters or login names in a regular rhythm, then going quiet and coming back with a narrower set. Neither CERT-EU nor the FBI publishes an ARTEX user agent, so do not rely on a user-agent string; agents can send anything.
- Egress. The FBI lists outbound calls to LLM APIs and to webhook.site from servers that should never make them as things to investigate, not block. A server in your pre-production network that starts talking to a model API is worth a look.
Credentials on customer portals. Bell's stolen data is a phishing kit for its customers' portals. Long-lived passwords on enterprise portals, shared team logins, and API keys with no expiry are what an agent, or a person with the agent's output, tries next. MFA on every portal account, unique credentials per person, short-lived API tokens and an alert on logins from new locations are the controls that still hold when the attacker is fast.
What our data shows
- Indicators: neither primary publishes any. Bell is sharing indicators privately with customers and the Cyber Centre; the Commission and CERT-EU published none. We fed nothing for these two incidents: none published in the sources we read.
- ARTEX: our index already holds ARTEX infrastructure from earlier public reports, including the 22 addresses in the FBI FLASH. All of it came from other people's published research. That is coverage, not a lead, and none of it is tied to the Commission or Bell incidents by any source.
- Maximo at our edge: we searched our edge records since September 1 for request paths containing /maximo. Across 727,791 edge blocks, 165,296 canary records and 2,477 origin honeypot records, there were zero. We run no Maximo, so this says only that nobody probed our sites for it; it says nothing about Bell or about you.
- No first-party sighting of either incident.
Sources
- BleepingComputer, "Bell Canada says 'agentic' hacker stole enterprise customer data," November 20, 2026: https://www.bleepingcomputer.com/news/security/bell-canada-says-agentic-hacker-stole-enterprise-customer-data/
- The Globe and Mail, "Bell says enterprise customer data was stolen in 'agentic' cybersecurity incident": https://www.theglobeandmail.com/business/article-bell-cybersecurity-incident-customer-data/
- SC World brief (returned HTTP 403 to us; not read): https://www.scworld.com/brief/bell-canada-confirms-data-breach-involving-agentic-threat-actor
- Reuters via The Star, "EU Commission confirms it was targeted in cyberattack through use of AI tools," November 19, 2026: https://www.thestar.com.my/tech/tech-news/2026/11/19/eu-commission-targeted-in-cyberattack-politico-reports
- The Record (Recorded Future News), November 19, 2026: https://therecord.media/european-commission-confirms-cyberattack-using-ai
- The Register, November 19, 2026: https://theregister.com/cyber-crime/2026/11/19/european-commission-scrambles-to-contain-ai-assisted-cyberattack/5311103
- CERT-EU, "From Penetration Testing to Cyber Threat: Beware of ARTEX," November 5, 2026: https://cert.europa.eu/blog/from-penetration-testing-to-cyber-threat-beware-of-artex
- FBI, FLASH-20261113-001, November 13, 2026: https://www.ic3.gov/CSA/2026/261113.pdf
We could not read Politico's original Brussels Playbook item or a Commission press release (we found none). Everything attributed to Politico comes through Reuters, The Record and The Register.
Five percent of this is wrong somewhere. It always is.
Was this useful? Rate this post. The widget is at the bottom of the page, and we read every response.
Every indicator in this post is in the feed. Free.
1.58M+ IOCs, STIX 2.1 / TAXII, 88% novel vs ThreatFox, exploited-CVE leads ahead of CISA. No credit card — a free API key in 30 seconds, and you can audit every claim above against the live endpoints.