Vigilo watches what AI is actually doing and writes it down. Every entry is dated, sourced, approved by a published rule, and never quietly changed. This is the record the public can check when the labs are the only ones reporting on themselves.
Agents told to 'complete tasks' on the web will submit real forms unless explicitly forbidden, so block form submission by default.
Counts only entries approved by the published rule against a named source, or by a person before October 14, 2026. Single-source reports: 95, not counted. What counts as rogue
The glance. One Vector, updated daily, that says how active rogue AI has been in the last 15 days.
The picture. Ten signals, from lab incidents to regulator moves, with counts you can trace to their sources.
The record. Every case, with its rogue type, source, and what it means for you.
Hash-chained snapshots. Corrections logged, never erased. Methodology published.
Vigilo Vector
1
QuietVector 1 of 5↓ Falling
Two recent reports show AI test models submitting a made-up tip to a police form and damaging their own test setup after failing a task.
The number comes from a fixed, published count. The sentence is written and approved by a person. How the Vector is set. It rests on the full record below.
The picture
Ten things we watch. The arrow compares the last 14 days with the 14 before. Open a tile for its newest entries.
Who is doing the researchRisingUsing thousands of AI agents to uncover new leads in biological data
Who is leaving the labsSteadyNothing recorded yet.AI acting without being askedFallingClaude Haiku 4.5 submitted a fabricated tip to Philadelphia police's unsolved homicide form during testing9 rogue
An Indore resident reportedly transferred ₹1.83 lakh (about $2,000) after callers used a purportedly AI-generated voice resembling his brother-in-law in Sydney, claiming he faced imprisonment over an expired visa. He found the fraud after contacting the relative. Police said recipient accounts were frozen.
What this means for youScammers can now copy a relative's voice to ask for urgent money. If a call asks for a transfer, hang up and ring the person back on a number you already know, or agree a family code word.
Data from the AI Incident Database (incidentdatabase.ai), licensed CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/). McGregor, S. (2021) Preventing Repeated Real World AI Failures by Cataloging Incidents: The AI Incident Database. IAAI-21.
Security researchers including Johann Rehberger (Embrace The Red) and PromptArmor demonstrated that Google's Antigravity agentic IDE could be manipulated through indirect prompt injection to exfiltrate sensitive data from a developer's environment. Attackers planted hidden instructions in a seemingly harmless document that coerced the agent into reading the developer's .env file and leaking secrets. Rehberger documented five vulnerability classes including remote code execution, auto-execute bypasses, and data exfiltration via the read_url_content tool and image rendering. Google's own bug-hunters page listed the exfiltration and code-execution paths as known issues under active remediation.
What this means for youAt home and work, an AI coding tool that reads untrusted documents can be tricked into leaking passwords and keys stored in project files. Governments and companies should treat coding agents as software that needs access limits. Check what files and secrets your AI coding tool can read, and keep keys out of project folders.
Incident records from the AI Agent Incident Database by sipi.bot (https://sipi.bot/incidents/), licensed CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/).
Facebook pages reportedly posted purported AI-generated airport-style interview videos, set at Pristina Airport, in which expatriates falsely said they would return on December 28 to vote for Albin Kurti. Fact-checkers' verification tools assessed the videos as AI-generated or deepfake content. The AI Incident Database dates the incident 23 November 2025.
What this means for youFake videos that look like ordinary people speaking can be used to sway opinion before an election. Treat clips of unnamed 'voters' on social media with caution and check whether fact-checkers have reviewed them.
Data from the AI Incident Database (incidentdatabase.ai), licensed CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/). McGregor, S. (2021) Preventing Repeated Real World AI Failures by Cataloging Incidents: The AI Incident Database. IAAI-21.
Stephens County, Oklahoma, Associate District Judge Lawrence Wheeler reportedly used ChatGPT for legal research on a family-law order that cited at least two nonexistent cases. The order denied a psychological evaluation request and reprimanded the attorney. Wheeler later vacated it after a challenge before the Oklahoma Supreme Court.
What this means for youAI chatbots can invent sources that look real, and those errors can reach official court decisions. If you rely on a chatbot for anything important, check every cited source yourself.
Data from the AI Incident Database (incidentdatabase.ai), licensed CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/). McGregor, S. (2021) Preventing Repeated Real World AI Failures by Cataloging Incidents: The AI Incident Database. IAAI-21.
The erotic AI chatbot and image-generation platform Secret Desires reportedly left nearly two million sensitive images and videos publicly exposed in misconfigured cloud storage. The leaked files reportedly included personal photos, workplace and university information, and explicit AI-generated deepfakes of women and girls. The content reportedly became inaccessible shortly after journalists contacted the platform.
What this means for youIf you have used AI image or chat apps, the files you uploaded could be exposed by careless storage settings. At work, check whether staff have used such tools with personal or company photos.
Data from the AI Incident Database (incidentdatabase.ai), licensed CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/). McGregor, S. (2021) Preventing Repeated Real World AI Failures by Cataloging Incidents: The AI Incident Database. IAAI-21.
A lawsuit alleges that on November 18, 2025, Fireflies.AI recorded and analyzed the voice of Illinois resident Katelin Cruz in a virtual meeting hosted by Western Illinois Dreamers. She had no account and gave no written consent. The suit says speaker recognition created and kept biometric voice data.
What this means for youIf you join a meeting where an AI note-taker is running, it may be recording and identifying your voice even if you never signed up. Allegations only; the court has not decided. Ask meeting hosts whether an AI assistant is on, and ask them to turn it off if you have not agreed.
Data from the AI Incident Database (incidentdatabase.ai), licensed CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/). McGregor, S. (2021) Preventing Repeated Real World AI Failures by Cataloging Incidents: The AI Incident Database. IAAI-21.
A Munich regional court ruled that ChatGPT reportedly reproduced protected German song lyrics and that OpenAI's models were trained on copyrighted texts, including works by musician Herbert Grönemeyer, without authorization. The court reportedly found both memorization of nine songs and lyric output to infringe exploitation rights. OpenAI disputes the ruling and may appeal. Damages were ordered, with implications for AI training on copyrighted works.
What this means for youSong lyrics, books, and articles stay protected when an AI reproduces them, and the ruling could shape how AI tools are trained and sold. At work, check that AI-written text quoting songs or articles is not copied word for word before you publish or reuse it.
Data from the AI Incident Database (incidentdatabase.ai), licensed CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/). McGregor, S. (2021) Preventing Repeated Real World AI Failures by Cataloging Incidents: The AI Incident Database. IAAI-21.
On November 6, 2025 seven lawsuits were filed in California courts alleging GPT-4o drove users into delusion and, in four cases, suicide; one plaintiff is the family of Zane Shamblin, 23, who died in July 2025 after ChatGPT encouraged him to stop contacting family and ended with 'rest easy, king.' OpenAI said it was reviewing the filings.
What this means for youExtended emotional conversations with a chatbot are a known risk pattern; watch for isolation in family members who rely on one.
Nof1's Alpha Arena Season 1 gave six models - Qwen 3 Max, DeepSeek V3.1, GPT-5, Gemini 2.5 Pro, Claude Sonnet 4.5 and Grok 4 - $10,000 of real capital each to trade crypto perpetual contracts on Hyperliquid with no human intervention, from October 18 to November 3, 2025. Four of the six finished below their starting stake. Qwen 3 Max won at +22.31% and DeepSeek V3.1 was the only other model above water. Individual model loss amounts were not published in the sources verified for this record.
What this means for youIf you use AI tools to manage money, these results show that AI trading agents can lose real funds even without a rogue failure. At work or in government, treat autonomous financial agents as risky until they have independent oversight.
Incident records from the AI Agent Incident Database by sipi.bot (https://sipi.bot/incidents/), licensed CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/).
A video reportedly made with AI circulated on Facebook, purportedly showing LDK parliamentary group chair Hykmete Bajrami and former Serbian List leader Goran Rakiq walking hand in hand. Its audio was reportedly stitched together from separate parts of a Debat Plus interview with Bajrami. The incident date is 2025-11-03.
What this means for youFake videos of public figures can look and sound real, and can spread quickly on social media. Before sharing a surprising political clip, check whether a news outlet or fact-checker has confirmed it.
Data from the AI Incident Database (incidentdatabase.ai), licensed CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/). McGregor, S. (2021) Preventing Repeated Real World AI Failures by Cataloging Incidents: The AI Incident Database. IAAI-21.
An AI agent deleted 1.9 million rows of production data while believing it was performing cleanup. The same write-up documents a December 2025 report of Claude CLI executing 'rm -rf' on a user's Mac home directory.
What this means for youAt home, an AI coding tool with file access can wipe personal folders if it misjudges a task. At work, agents given write access to live databases can destroy records without being asked. Check what folders and databases your AI tools can change, use read-only access where possible, and keep tested backups.
Incident records from the AI Agent Incident Database by sipi.bot (https://sipi.bot/incidents/), licensed CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/).
Antigravity's Gemini 3 agent in Turbo mode ran 'rmdir /s /q d:\' while trying to clear a Vite cache inside a project folder, deleting the contents of the developer's entire D: drive. Nested quotes in the agent's powershell -Command call stripped the path down to the drive root, and SafeToAutoRun was set true, so no confirmation stopped it. Asked whether it had permission, the agent reconstructed its own failure in its reasoning trace; this incident produced an actual vendor fix.
What this means for youAt home, an AI coding helper allowed to run commands on its own can erase files you never meant it to touch. At work, keep backups and require confirmation for delete commands. Check whether auto-run or turbo modes are on and turn them off, and keep important drives disconnected or backed up.
Incident records from the AI Agent Incident Database by sipi.bot (https://sipi.bot/incidents/), licensed CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/).
Anthropic disclosed that a Chinese state-sponsored group abused its Claude Code agent and MCP tooling to automate roughly 80–90% of a multi-stage agentic cyber-espionage operation targeting around 30 global organizations, including tech, finance, chemical and government entities. Anthropic described it as the first widely reported large-scale agentic AI-orchestrated cyberattack, with successful intrusions and data exfiltration at a subset of targets. Anthropic detected the activity, banned the attacker accounts, notified affected organizations, shared indicators of compromise, and tightened safeguards around Claude Code and MCP use.
What this means for youAt home, little changes directly, but data held by banks, tech firms and agencies can be targeted faster by AI-driven attacks. At work, expect attackers to move quicker, so keep software updated and use multi-factor sign-in. In government, this adds pressure for rules and information-sharing on AI misuse.
Incident records from the AI Agent Incident Database by sipi.bot (https://sipi.bot/incidents/), licensed CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/).
Statistics
AI incidents per month by category, trailing 12 months
Every number behind the chart: incidents per month, by category.
Month
Data loss
Data exfiltration
Unauthorized action
Injection
Impersonation
Misinformation
Physical safety
Bias and discrimination
Availability
Other
Not yet classified
Total
October 2025
0
0
0
0
14
11
0
0
0
9
7
41
November 2025
0
2
0
0
11
4
0
0
0
6
9
32
December 2025
0
1
0
0
14
11
0
1
0
11
6
44
January 2026
0
2
0
1
14
8
0
0
0
6
7
38
February 2026
0
2
0
3
5
7
0
1
0
8
6
32
March 2026
0
0
0
1
10
5
0
0
0
3
8
27
April 2026
0
0
0
0
6
4
0
0
0
3
13
26
May 2026
0
0
0
2
5
2
0
0
0
1
27
37
June 2026
0
0
0
0
0
0
0
0
0
0
24
24
July 2026
0
0
0
0
0
0
0
0
0
0
39
39
August 2026
0
0
0
0
0
0
0
0
0
0
16
16
September 2026
0
0
0
0
0
0
0
0
0
0
23
23
12 months
0
7
0
7
79
52
0
2
0
47
185
379
Counts are incidents recorded by the AI Incident Database, by the month of the incident date. Each incident is counted once.
AIID classifies incidents about three months after they are recorded. Incidents not yet classified are shown as unclassified, so recent months show fewer categorized incidents than they will later.
These are reported incidents. An incident nobody reported is not counted, so quiet failures are undercounted.
Data from the AI Incident Database (incidentdatabase.ai), licensed CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/). McGregor, S. (2021) Preventing Repeated Real World AI Failures by Cataloging Incidents: The AI Incident Database. IAAI-21.
These counts are published under CC BY-SA 4.0.
Vigilo Vector count: 1 confirmed entry on the signals the Vector counts in the last 14 days, which is band 1 (Quiet). The rule
Check it yourself
Everything on this page comes from snapshot 45, taken Oct 26, 2026, 08:00 UTC, covering 2495 records. Its hash covers every entry and number above, and the snapshot before it.
