Monitoring sources · Refreshed Nov 17, 2026
Since your last visit Checking the verified record against your last visit on this device.

AI Attack News · AI Attack Tracker · Evidence Atlas

AI Attack News, tracked as evidence

News first, the tracker beside it, the map under the tracker. Verified reports only.

Verified reports 11 Latest publication Nov 16, 2026 Map pins 5 documented countries Listing rule Finder publishes first

AI Attack News

Latest AI attack news today · Updated Nov 17, 2026
Lead storyNov 16, 2026 · 1 day agoGoogle GTIGAgent run · High and Medium

Agentic threat activity crosses a significant threshold

The newest AI attack news is a step change in autonomy. Google GTIG reports exploit_agent, a system found on an exposed staging server that compares software before and after a patch ships, works out the flaw the patch closed, drafts an exploit and runs it with no human in the loop. The same report places the open source agent frameworks CAIRN and PentAGI inside live intrusions, describes MIDNIGHT NEPTUNE turning a coding assistant setup into a backdoor route into crypto targets, and updates LLMJacking, where actors break into cloud environments and burn the victim AI bills and quotas as the prize itself.

Targets Web software · US aerospace · Asia · CryptoTracker records 4 in this reportEvidence Provider reported and Confirmed

More headlines · newest first

Feb 25, 2026Gambit SecurityAgent run A single operator, two AI platforms, nine government agencies One operator used Claude Code and GPT 4.1 against nine Mexican government organisations. Gambit reconstructed the campaign from recovered logs counting 1,088 prompts and 5,317 AI generated commands, with AI running about 75 percent of remote actions. Several named agencies denied being breached, so the tracker grades this story partly contested and says so on the record. Feb 20, 2026Amazon Threat IntelligenceAI assisted AI augmented threat actor accesses FortiGate devices at scale Amazon reports one actor assessed as low to medium skill, significantly augmented by AI at every phase, reaching 600 plus FortiGate firewalls across 55 plus countries. Entry came through exposed management ports and weak credentials, not a firewall flaw. The tracker grades the scale critical and the finding confirmed. Nov 13, 2025AnthropicAgent run Disrupting the first reported AI orchestrated cyber espionage campaign Anthropic describes GTG 1002 as the first reported AI orchestrated espionage campaign, assessed as Chinese state sponsored, aimed at roughly 30 global targets with AI performing an estimated 80 to 90 percent of the campaign. The same disclosure covers GTG 2002, extortion run through Claude Code against 17 organisations. Nov 5, 2025Google GTIGAI in malware Advances in threat actor usage of AI tools This GTIG report introduced two malware stories that still define the category. PROMPTSTEAL, used by Russian APT28 against Ukraine, asks a language model what commands to run while it operates. PROMPTFLUX asks Gemini to rewrite and obfuscate itself, in one variant every hour, and GTIG places it still in testing with no compromise of a network or device shown. Analysis · Q2 2026 activityGoogle GTIGBackground From prompting to autonomy: the evolution of adversarial AI The background read for the whole AI attack news page. GTIG traces the move from attackers prompting models for help to agents planning and running work on their own, including a Q2 2026 case where attackers broke into a cloud resource and then planned, built and ran a mass credential harvesting campaign with agents in under six hours.

AI Attack Tracker

11 records · newest first
Showing 11 / 11 records · tap a row for detail
NOV 16, 2026 · 1 day agoDelay not statedNewHigh

exploit_agent: exploits built and fired with no human in the loop

Found on an exposed staging server. Compares software before and after a patch ships, derives the flaw the patch closed, drafts an exploit and runs it on its own. Autonomy: fully autonomous, per GTIG.

Activity not dated in report
Popular web software · worldwideAgent runProvider reportedGTIG ↗
NOV 16, 2026 · 1 day agoDelay ≈1 moNewHigh

CAIRN and PentAGI: agent frameworks inside live intrusions

Open source multi agent frameworks automating phases of real intrusions. CAIRN seen against US aerospace targets. A customised PentAGI platform scanned organisations across Asia.

Activity Aug to Oct 2026
Aerospace sector · United States · Organisations · AsiaAgent runConfirmedGTIG ↗
NOV 16, 2026 · 1 day agoDelay not statedNewMedium

MIDNIGHT NEPTUNE: a coding assistant turned into a backdoor

Cluster GTIG suspects is North Korean. Fake recruiter profiles, then a rigged coding assistant setup as the channel in.

Activity not dated in report
Crypto sector · North America and EuropeAI assistedProvider reportedGTIG ↗
Q2 2026 · ≈5 mo agoDelay not statedNewHigh

Credential harvesting planned and run by agents in under six hours

Attackers broke into a cloud resource, then planned, built and ran a mass credential harvesting campaign with agents in under six hours. Same reporting covers supply chain tricks aimed at AI coding assistants.

Activity Q2 2026
Cloud sector · open source supply chains · worldwideAgent runProvider reportedGTIG ↗
FEB 25, 2026 · ≈9 mo agoDelay not statedNewHigh

One operator against nine Mexican government organisations

Claude Code and GPT 4.1. Recovered logs: 1,088 prompts, 5,317 AI generated commands, AI running about 75 percent of remote actions. Several named agencies denied being breached.

Activity not dated in report
Government sector · MexicoAgent runPartly contestedGAMBIT ↗
FEB 20, 2026 · ≈9 mo agoDelay not statedNewCritical

600 plus firewalls in 55 plus countries, one low skill actor

Actor assessed low to medium skill, significantly augmented by AI at every phase. Entry through exposed management ports and weak credentials, not a firewall flaw.

Activity not dated in report
Network infrastructure · FortiGate firewalls · 55 plus countriesAI assistedConfirmedAWS ↗
NOV 13, 2025 · 1 yr agoDelay not statedNewCritical

GTG 1002: the first reported AI orchestrated espionage campaign

Group assessed as Chinese state sponsored. Espionage against roughly 30 global targets, AI performing an estimated 80 to 90 percent of the campaign.

Activity not dated in report
Tech, finance, chemical, government sectors · worldwideAgent runProvider reportedANTHROPIC ↗
NOV 5, 2025 · ≈1 yr agoDelay ≈1 moNewHigh

PROMPTSTEAL: malware that asks an AI what to do next

Used by Russian APT28 against Ukraine. Instead of fixed commands, the malware asks a language model to generate them while it runs, then ships documents and system information out. First reported by CERT UA, Jul 2025.

Activity live use since Jun 2025
Government and civilian targets · UkraineAI in malwareConfirmedGTIG ↗
NOV 5, 2025 · ≈1 yr agoDelay not statedNewLow

PROMPTFLUX: malware that asks Gemini to rewrite itself

Dropper prompting the Gemini API for fresh obfuscation, in one variant rewriting its own source every hour. Still in testing, no compromise of any network or device shown, per GTIG.

Activity testing phase · not dated
None demonstratedExperimentalExperimentalGTIG ↗
AUG 2025 · ≈15 mo agoDelay ≤1 moNewHigh

GTG 2002: extortion run through Claude Code

One cybercriminal used Claude Code as the operational core of a data extortion campaign against 17 organisations.

Activity summer 2025
Healthcare, emergency services, government, religious sectors · 17 organisationsAI assistedProvider reportedANTHROPIC ↗
2026 · in 2026Delay not statedNewMedium

LLMJacking: hijacked cloud AI bills as a business model

Mandiant responded to several cases where actors broke into cloud environments and burned victim compute and model quotas for their own workloads. The AI is the loot, not the weapon.

Activity 2026 · exact dates not stated
Cloud sector · environments with exposed AI credentialsAI in malwareConfirmedGTIG ↗

AI Attack Map

Evidence Atlas · 5 pins
Evidence Atlas · Country level references only · Verified published reports
1 UKRAINE 2 MEXICO 3 UNITED STATES 4 SINGAPORE 5 UNITED KINGDOM
Documented country Autonomous evidence leads the pin New since your last visit · pulses once Documented connection · Russia to Ukraine, APT28, per GTIG

Listing rule: a country is pinned only when the published report documents that country, and a line is drawn only where a source report documents the connection. Reports that name no country get a tracker record and no pin.

Evidence Atlas · custom drawn vector map · coastlines simplified · Singapore anchors the Asia scanning GTIG describes, and the United Kingdom anchors its Europe targets. Tap a pin to highlight its tracker records.

Report Updates

Latest updates from the teams that published the reports · Tap through to the source

Update cards built only from the verified reports already in the tracker above. Each card opens the original report from the team that found the attack.

Google GTIGReportNov 16, 2026

Agentic threat activity crosses a significant threshold. exploit_agent builds and fires exploits with no human in the loop, and CAIRN and PentAGI agent frameworks were seen inside live intrusions.

Open original report ↗
AnthropicReportNov 13, 2025

Disrupting the first reported AI orchestrated cyber espionage campaign. GTG 1002, assessed as Chinese state sponsored, with AI performing an estimated 80 to 90 percent of the campaign.

Open original report ↗
Amazon Threat IntelligenceReportFeb 20, 2026

AI augmented threat actor accesses FortiGate devices at scale. One low to medium skill actor, augmented by AI at every phase, reaching 600 plus firewalls in 55 plus countries.

Open original report ↗
Gambit SecurityReportFeb 25, 2026

A single operator, two AI platforms, nine government agencies. Recovered logs counted 1,088 prompts and 5,317 AI generated commands. Several agencies denied being breached, so this one is graded partly contested.

Open original report ↗

Service Availability · Internet and Big Tech Status

Service availability only · Official status sources · Not attack reports

This board answers one question only: is the service up or down right now. It is a service availability board, not a list of attacks, and a degraded service here does not mean anyone was hacked. Snapshot checked Wed Nov 18, 2026 at 12:37 PM ET from official status feeds. Rows marked Live check try to refresh from the official status API when you open this page and show the time they updated. Rows marked Check official have no simple public feed in this snapshot, so tap through instead of trusting a stale label.

OpenAIChatGPT and API · Live check
Up · Operational
Snapshot Nov 18, 2026 · Official status ↗
AnthropicClaude · Live check
Degraded · Partially degraded service
Snapshot Nov 18, 2026 · Official status ↗
CloudflareNetwork and DNS · Live check
Degraded · Partially degraded service
Snapshot Nov 18, 2026 · Official status ↗
GitHubRepos and Actions · Live check
Up · Operational
Snapshot Nov 18, 2026 · Official status ↗
Google CloudCloud and Workspace
Up · No active incident in official feed
Snapshot Nov 18, 2026 · Most recent incident in feed ended Oct 20, 2026 · Official status ↗
Amazon AWSCloud infrastructure · Check official
Check official page
Feed checked Nov 18, 2026 · Official status ↗
MicrosoftAzure and Microsoft 365 · Check official
Check official page
No simple public feed in this snapshot · Official status ↗
MetaFacebook, Instagram, WhatsApp Business · Check official
Check official page
No simple public feed in this snapshot · Official status ↗

How to read it: circle shape plus Up means operational. Triangle shape plus Degraded means the official feed reported a partial problem at snapshot time. Square shape plus Check means open the official page for the current word. Colour is never the only signal, every pill also carries a shape and a word.

Data and API

Tracker data · JSON export
GET/api/v1/incidentsAll 11 records
GET/api/v1/incidents?type=agent_runFilter by AI role
GET/api/v1/incidents?status=confirmedFilter by confidence
GET/api/v1/incidents?severity=criticalFilter by severity
Last updated Nov 17, 2026

        

This static page has no server, so the export above is generated in your browser from the same dataset as the map and feed.

Learn and Reference

Education and background, kept separate from the live tools above and collapsed until you open it. The news, tracker and Evidence Atlas above are the main area.

About AI Attack News

Reference
About AI Attack News · Read more

AI Attack News is the news page of the AI Attack Tracker. If you searched for AI attack news, AI attack news today or latest AI attack news, this is the page that pulls the stories together in one place, newest first, with what happened, who found it, who was targeted and how serious the tracker grades it.

Every story here comes from a published report by the company or agency that found the attack. AI Attack News summarises that report, links you to the full original reporting, and links you back to the matching record in the AI Attack Tracker and on the AI Attack Map. No rumours, no unnamed claims and no story without a source report.

AI Attack News at aiattacknews.com collects the latest AI attack news on autonomous AI attacks, agentic AI attacks, AI assisted intrusions, AI in malware, AI powered credential harvesting and AI orchestrated espionage. It sits alongside the AI Attack Tracker and the AI Attack Map so a reader who arrives for AI attack news today can move straight to the evidence: the tracker record, the severity and evidence grade, the map pin where a location is documented, and the original published report. When a new verified AI attack report is published, it belongs in AI Attack News, in the tracker feed and on the map together.

Dates on this page are publication dates. Where a report gives a different activity date or a publication delay, the tracker record shows both, and where a source gives no activity date the record says not dated in report rather than guessing.

About the Tracker

Reference
What this tracker tracks · Read more

Use this AI attack tracker to track AI attacks worldwide: autonomous AI attacks, agentic AI attacks, AI assisted intrusions, AI malware and AI powered cyber attacks. The Evidence Atlas plots every verified incident with a documented country, and the AI Attack Tracker rail beside the news lists each AI attack with its source report, severity, AI role and evidence grade. If you are looking for an AI threat tracker, an AI incident tracker or a live AI cyber attack tracker, this is that tracker, built only from published reports.

The AI Attack Tracker at aiattacknews.com is a live AI attack tracker, AI threat tracker and AI incident tracker for the new wave of AI cyber attacks. The tracker follows autonomous AI attacks run by AI agents, agentic AI attacks, AI assisted intrusions, AI in malware, AI powered credential harvesting and AI orchestrated espionage. Every record in this AI cyber attack tracker is plotted on the AI Attack Map when a location is documented and listed in the tracker feed, with the original published report linked on each record. AI Attack News is the news companion to the tracker: it turns those same verified records into readable AI attack news stories for readers searching for AI attack news and AI attack news today. Search the tracker by AI role, severity and evidence grade, download the AI attack tracker data as JSON, read the latest stories in AI Attack News, and follow the AI attack disclosure timeline to see when each AI attack was found and published.

AI Attack Tracker is the live AI attack tracker and AI attack map on aiattacknews.com, with AI Attack News for the latest AI attack news today. The AI Attack Map domain aiattackmap.com and the AI Attack Tracker domain aiattacktracker.com redirect to aiattacknews.com. Only published reports.

Timeline

Education · Publication dates
Show disclosure timeline
NOV 16, 2026Agentic threat activity crosses a significant threshold · exploit_agent, CAIRN and PentAGI, MIDNIGHT NEPTUNE, LLMJacking updateGoogle GTIG
FEB 25, 2026Single operator against nine Mexican government organisationsGambit Security
FEB 20, 2026AI augmented actor accesses FortiGate devices at scaleAmazon Threat Intelligence
NOV 13, 2025First reported AI orchestrated espionage campaign · GTG 1002 and GTG 2002Anthropic
NOV 5, 2025Advances in threat actor usage of AI tools · PROMPTSTEAL and PROMPTFLUXGoogle GTIG

How attacks were detected

Education · Methods from source reports
Show detection notes

Provider telemetry

Google GTIG and Anthropic identified activity through model use signals, account investigation, and patterns in prompts and agent actions described in their own reports.

Incident response

Amazon Threat Intelligence and Mandiant cases came from investigations of compromised cloud and network environments, with findings published by the responding organisation.

Recovered artefacts

Gambit Security reconstructed the Mexico campaign from recovered logs, including prompt and command counts quoted in its technical report. Contested points are labelled in the feed.

Grading key

Reference
Show grading key
CONFIDENCE · Confirmed = corroborated beyond the finder · Provider reported = published by the finder only · Partly contested = named targets dispute part of the report · Experimental = capability shown, no compromise demonstrated.
SEVERITY · Tracker grade from scope and demonstrated impact in the source report · Critical 4 bars · High 3 · Medium 2 · Low 1 · Delay = activity date to publication date, or not stated where the source gives no activity date.