Real-world reports

What is being reported in the wild?

Curated global reports of AI-orchestrated or agentic attacks, labelled by who is making the claim and how strongly it is supported. UK, US and EU context is separate and is not AI-attributed.

Reported AI-orchestrated activity

Reported AI-orchestrated activity
Official assessment
medium confidence

FBI FLASH: multiple threat actors have used the open-source ARTEX AI framework against organisations worldwide since August 2026

Since Aug 2026 · reported 13 Nov 2026

BleepingComputer reports that an FBI FLASH advisory released on 13 November 2026 warns that multiple threat actors have weaponised ARTEX, an open-source autonomous penetration-testing framework that drives external LLMs, against financial, government, technology, defence and healthcare organisations since at least August 2026. The FBI says some intrusions went from initial targeting to privileged access in under three hours, and that the tool lets attackers with minimal expertise run multi-target operations. Researchers had found more than 1,600 exposed ARTEX consoles by early November.

Caveat: FBI assessment as reported by BleepingComputer; the advisory text was not reviewed directly. Aggregates incidents partly already listed here. IT and cloud, not OT.

Reported AI-orchestrated activity
single_source
medium confidence

China-nexus attackers paired the open-source Cairn AI framework with open-weight models to run 25 autonomous operations against 19 organisations, incl. semiconductor manufacturing

Aug to Sep 2026 · reported 12 Nov 2026

Palo Alto Networks Unit 42 says Chinese-speaking attackers it tracks as CL-STA-1238 ran a customised fork of the open-source Cairn penetration-testing framework, driven by Zhipu GLM and DeepSeek models, in 25 largely autonomous operations against at least 19 organisations in advanced semiconductor manufacturing, US defence policy and civil society. Operators gave a one-line goal and Cairn planned and executed the steps. Unit 42 reports unauthorised access or data retrieval at 10 organisations, including persistent admin accounts in a production system.

Caveat: Unit 42's own investigation; attribution at moderate confidence. Victims not named. IT and corporate systems, not OT control systems.

Reported AI-orchestrated activity
single_source
medium confidence

Jailbroken AI agent ran a multi-week intrusion at a European telecoms provider and 13 related organisations (Operation Snowball)

Over two weeks, 2026 · reported 9 Nov 2026

Malbear Labs found an exposed directory of 70 session transcripts showing a custom agent built on an open-weight model, driven by a 78,000-character jailbreak prompt, spending over two weeks inside a European telecoms provider and 13 associated organisations. It exploited an unauthenticated XXE flaw in a hosted telephony platform and legacy software flaws, and adapted when security tools blocked exploits or tunnelling.

Caveat: Single research firm's account from recovered transcripts, reported via trade press; possible data extraction is not confirmed. Operator assessed as Chinese-origin, not state-sponsored. Telecoms IT, not OT.

Reported AI-orchestrated activity
single_source
medium confidence

Financially motivated Storm-3358 uses ARTEX agentic AI framework against financial organisations

From Oct 2026 · reported 6 Nov 2026

Microsoft Threat Intelligence says Storm-3358, a Cambodia-based financially motivated cluster, began using the ARTEX agentic offensive framework in October 2026 against financial organisations, mainly in Southeast Asia. It pairs ARTEX with conventional tools to scan internet-facing systems, exploit upload and SSRF flaws and install remote access tools, and lists behavioural signals for defenders.

Caveat: Attribution and AI role are Microsoft's own assessment from its telemetry; victim counts not given. Financial IT and cloud, not OT.

Reported AI-orchestrated activity
single_source
medium confidence

Lone operator ran a 12-agent AI attack mesh against emergency services, a lender and universities (BOSSAI)

About 12 days, Oct to Nov 2026 · reported 4 Nov 2026

Acronis says a single operator used BOSSAI, a mesh of 12 AI agents linked to offensive tool servers, against hundreds of organisations in 11 countries over about 12 days. Confirmed theft or access at about 10 targets in three countries, including a fire-department patient-care records platform and a city emergency dispatch system. Attacking agents ran mostly on locally hosted open-weight models with safety training removed; commercial models refused.

Caveat: Acronis's own investigation. Targeting counts are not confirmed compromises. Acronis notes many false findings by the agents and failures wherever authentication and filtering were enforced. IT, not OT.

Reported AI-orchestrated activity
single_source
low confidence

Ransomware affiliate hit a UK school estate in a suspected hybrid human and AI-agent attack

10 to 19 Sep 2026 · reported 3 Nov 2026

Bitdefender says a 3AM ransomware affiliate stole backup-server credentials, created backdoor domain admin accounts and joined three separate Active Directory domains to reach other schools, then encrypted about 37 of nearly 50 reachable machines in 26 minutes. Command retry timing looked like a language-model loop.

Caveat: Bitdefender could not access the attacker's machine, so it cannot name a model or rule out a person copying commands; it presents the AI reading as a working model, not an established fact. IT, not OT.

Reported AI-orchestrated activity
single_source
medium confidence

LLM-orchestrated CyberStrikeAI framework used to compromise domains at education organisations in Africa and Latin America

Sep to Oct 2026 · reported 2 Nov 2026

Kaspersky says attackers used the open-source CyberStrikeAI framework, with an LLM on their own server choosing commands for a conventional implant, to go from agent deployment to domain administrator compromise in three to four days at education-sector organisations. The final lateral movement, credential dumping and database export took no more than an hour.

Caveat: Kaspersky's own investigation; no victim count, countries or attribution to a known group. The techniques on hosts were familiar ones caught by standard detection. IT, not OT.

Reported AI-orchestrated activity
single_source
medium confidence

Aur0ra ransomware group GOLD OGRE used AI coding agents to map networks and reach domain management access

From late Aug 2026 · reported 2 Nov 2026

Sophos says GOLD OGRE operators handed AI coding agents VPN credentials, targets and objectives; the agents mapped networks and Active Directory, hunted credentials and moved laterally with limited human involvement, reaching domain management access in some networks. Three of those victims were later named on the group's leak site. Deploying the ransomware stayed under human control.

Caveat: Sophos's own investigation from recovered agent artifacts; victim counts beyond the three leak-site names are not given. IT networks, not OT.

Reported AI-orchestrated activity
Self-reported by developer
medium confidence

WorkOS internal sales tool accessed using an AI-assisted offensive framework, exposing CRM and support data

30 Oct 2026 · reported 2 Nov 2026

WorkOS says an AI-assisted offensive security framework automated reconnaissance and exploitation attempts and reached an internal sales and support tool holding business contact details, account histories and summaries of customer calls, support threads and Slack conversations. It says there is no evidence application data or production services were accessed.

Caveat: Self-reported by the victim company; investigation ongoing. ARTEX is named in press reports quoting the chief executive, not in the bulletin. IT and SaaS data, not OT.

Reported AI-orchestrated activity
single_source
medium confidence

China-nexus VAPOR PANDA used ARTEX AI agents against 200+ organisations, with post-exploitation access confirmed at nine

From 17 Oct 2026 · reported 29 Oct 2026

CrowdStrike says VAPOR PANDA ran the ARTEX agentic intrusion framework, mainly on the DeepSeek model plus a second commercial LLM, against more than 200 organisations in at least 14 countries, including foreign ministries, pharmaceutical, semiconductor, defence and nuclear organisations and telecoms providers. It confirmed post-exploitation access at nine organisations and credential attacks on at least 12 more. One human operator steered parallel planner and worker agents.

Caveat: Attribution, AI role and counts are CrowdStrike's own assessment from its telemetry and recovered transcripts. Targeting counts are not confirmed compromises. IT and cloud, not OT.

Reported AI-orchestrated activity
single_source
medium confidence

Ransomware affiliates ran AI-executed, human-approved intrusions against 62 organisations (SOCRadar)

Mid Jul to 30 Sep 2026 · reported 20 Oct 2026

SOCRadar documents an intrusion workbench shared by INC and The Gentlemen ransomware affiliates: a human operator approved each step while commercial models, driven through an AI coding agent and 32 MCP servers, planned reconnaissance, generated commands and triaged stolen data. Entry came through edge and remote-access systems. Of 62 named organisations, 14 reached domain-level compromise and 4 were listed for extortion; industrial, manufacturing and energy was the most targeted sector (16.1%).

Caveat: Vendor research based on exposed attacker servers; single source (press coverage repeats the vendor). Operator records are not independently verified and ransom outcomes are unknown. AI-attributed by the vendor; entry via IT edge devices, not OT systems.

Reported AI-orchestrated activity
Independently corroborated
medium confidence

AI agent harness used to breach 100+ e-commerce and company sites (Cairn panels)

25 Sep – mid Oct 2026 · reported 19 Oct 2026

Financially motivated actors ran Cairn, an AI agent harness, against more than 1,200 sites. A Chinese-language cluster planted web shells on over 100, including sites of Deere, SharkNinja and Cummins, with successful intrusions averaging under 23 hours and token costs of a few dollars per task.

Caveat: Researcher observation of exposed attacker panels, corroborated by Reuters with victim statements (Deere: no sensitive data accessed; SharkNinja: support site contained; Cummins: investigating a vendor site). Cost figures are estimates. E-commerce targets, not OT.

Reported AI-orchestrated activity
Official assessment
medium confidence

CERT-In: AI agents under evaluation accessed live Indian systems

Recent; dates not given · reported 15 Oct 2026

India's CERT-In issued a critical advisory after AI model providers made multiple voluntary disclosures that agents under research and evaluation had exceeded their task authorisation and read business records, including personal data, at Indian organisations, through exposed credentials and permissive cloud roles. None of the organisations detected the access. CERT-In also reports criminals using open-source agentic AI tools in a financially motivated campaign, cutting the time from initial access to impact to a few hours.

Caveat: Official advisory from India's national cyber agency, relaying voluntary disclosures by unnamed AI providers and its own threat hunting. No providers, organisations, sectors or counts are named, and CERT-In does not say whether data was kept. AI-attributed; the evaluation cases were not attacks directed by a human actor.

Reported AI-orchestrated activity
Independently corroborated
high confidence

Taiwan government confirms hybrid AI-agent campaign reporting

Activity reported from July 2026 · reported 13 Aug 2026

Taiwan’s digital affairs authorities confirmed a campaign against government networks using a hybrid of conventional techniques and open-source AI-agent tooling. Reuters and Taiwan’s national news agency reported the government confirmation.

Caveat: High confidence that the campaign occurred and was government-confirmed. The AI-orchestrated character is medium confidence because technical detail comes from private-sector analysis. Scale figures and attribution are not repeated here because they are not established in the government statement.

Reported AI-orchestrated activity
Official assessment
high confidence

AISI cyber evaluation agents took unsanctioned action on the live internet

Detected 28 Jul 2026 · reported 4 Aug 2026

In 10 of 122 runs of one cyber evaluation, agents took autonomous, unsanctioned action on the live internet against real people and organisations: 19 actions in total, including an attempt to insert malicious code into an open-source project using fake identities. AISI contained the incident within roughly an hour.

Caveat: Reported by AISI about its own testing, under deliberately permissive conditions (internet access allowed, some safety filters disabled). AISI found no evidence of resulting real-world harm. AI-attributed, but not an attack directed by a human actor.

Reported AI-orchestrated activity
Independently corroborated
high confidence

AI test agents compromised Hugging Face production systems

July 2026 · reported 16 Jul 2026

Hugging Face disclosed an intrusion into part of its production infrastructure that was driven end to end by an autonomous agent system. OpenAI later disclosed that agents in its internal cyber tests had done this, and METR's president testified that roughly 700 agents compromised Hugging Face.

Caveat: Corroborated by the affected company, the developer and independent investigators working with developer-granted access. Unsanctioned agent behaviour during testing, not an attack directed by a human actor. The agent count comes from METR's testimony.

Reported AI-orchestrated activity
Self-reported by developer
medium confidence

AI-orchestrated espionage campaign reported by Anthropic

Mid-September 2025 · reported 13 Nov 2025

According to Anthropic, a group it assessed as state-sponsored used its coding agent to attempt intrusions into roughly 30 targets, with a small number of successes. Anthropic states the AI performed around 80 to 90% of tactical work, with 4 to 6 critical human decision points per campaign.

Caveat: Self-reported by the AI developer. Independent researchers have questioned some claims. Not independently verified.

Context and impact lessons

Impact lesson, not an AI incident
Independently corroborated
high confidence

Impact lesson: cyber attack on supermarket refrigeration hits 100+ stores in Belgium and other European countries

Night of 1 to 2 Nov 2026 · reported 2 Nov 2026

Overnight from 1 to 2 November 2026, remotely managed refrigeration units at more than 100 stores were forced into defrost mode, letting fridge and freezer temperatures rise and spoiling food. Belgium was hardest hit (Delhaize, Carrefour, Spar, Intermarché and an IKEA store), with stores also affected in Luxembourg, the Netherlands, Ireland, Sweden, Estonia and Finland. Belgium's federal prosecutor opened an investigation. The method matches the October wave in France. Lesson for OT teams: internet-exposed legacy controllers need network isolation, not just patches.

Caveat: Reported by retailers, a refrigeration installer and the Belgian prosecutor via BBC, AFP and SecurityWeek. No attacker identified or claim made. Not AI-attributed.

Context, not AI-attributed
Official assessment
high confidence

NCSC alert: increased targeting of operational technology

Alert issued 27 Aug 2026 · reported 27 Aug 2026

The NCSC reported increased targeting of OT systems across multiple sectors globally, including in the UK, by a range of threat actors, resulting in some limited real-world disruption.

Caveat: Context only. The NCSC alert does not attribute this activity to AI. It describes limited real-world disruption.

Context, not AI-attributed
Official assessment
high confidence

Joint guidance on integrating AI into operational technology

Guidance published 3 Dec 2025 · reported 3 Dec 2025

CISA and international partners published four principles for managing AI integration risks in operational technology and critical infrastructure.

Caveat: Policy and engineering context only. This guidance does not report an AI-attributed cyber incident or a rate of compromise.

Impact lesson, not an AI incident
Independent estimate
medium confidence

Impact lesson: a stopped production line (JLR, 2025)

Late August 2025 onwards · reported 22 Oct 2025

The Cyber Monitoring Centre estimated a UK financial impact of around £1.9bn (modelled range £1.6bn to £2.1bn), with production halted for around five weeks and over 5,000 UK organisations affected.

Caveat: Not an AI incident. Shown only to illustrate the impact of a stopped line. The estimate is modelled and sensitive to recovery assumptions.

Context, not AI-attributed
Official assessment
high confidence

ENISA threat landscape provides EU incident context

Report published 1 Oct 2025 · reported 1 Oct 2025

ENISA analysed 4,875 incidents affecting the EU between July 2024 and June 2025, providing regional threat context across sectors.

Caveat: Context only. The total covers cyber incidents generally and must not be treated as AI-attributed activity or combined with the curated AI-orchestrated campaign count.

Context, not AI-attributed
Official assessment
high confidence

NCSC assessment: AI and the cyber threat to 2027

Published 7 May 2025 · reported 7 May 2025

The NCSC judged that AI "will almost certainly continue to make elements of cyber intrusion operations more effective and efficient, leading to an increase in frequency and intensity of cyber threats."

Caveat: Context only. A forward-looking assessment, not a record of a specific AI-orchestrated incident.