Vigilo Index
↑ rising
+1 in 24h
+10 in 7 days
From 9 counters built from the record. How this is calculated
Scale
- Model releases reported · this year↑rising
Verified entries on signal 9, model and agent releases.
Sources: Vigilo AI Watch public record · last updated Oct 22, 2026
- Models trained at GPT-4 scale est. · latest figure→steady
Estimate
Epoch AI's count of publicly disclosed models it estimates were trained with at least 10^25 FLOP, the scale of GPT-4, held until Epoch updates it. Nobody publishes a count of training runs underway.
- 33, Nov 21, 2025, Epoch AI, Models trained on over 1e25 FLOP
Sources: Epoch AI, models trained on over 10^25 FLOP · last updated Nov 21, 2025
- 2 counters awaiting data
- AI agents running worldwide (no published figure yet)
- Businesses adopting AI agents (no published figure yet)
Capability
- Benchmark records broken reported · this year↑rising
Verified entries classified as a new state-of-the-art result on a public benchmark.
Sources: Vigilo AI Watch public record; Epoch AI · last updated Oct 20, 2026
- Self-improvement papers published reported · this year↑rising
Verified entries on signal 1, lab research automation: AI doing the research that builds AI.
Sources: Vigilo AI Watch public record · last updated Oct 21, 2026
- 2 counters awaiting data
- Autonomous task milestones (not counted yet)
- Agent-to-agent deployments reported (not counted yet)
Harm
- Deepfake incidents reported · this year→steady
Verified entries in the impersonation category: deepfakes and cloned voices or likenesses.
Sources: Vigilo AI Watch public record; AI Incident Database · last updated Sep 10, 2026
- Jailbreaks disclosed reported · this year↓falling
Verified entries in the injection category: jailbreaks and prompt injection.
Sources: Vigilo AI Watch public record; AI Incident Database · last updated Oct 8, 2026
- Weaponized AI and malicious use reported · this year↓falling
Verified entries on signal 10: people deliberately using AI to harm others.
Sources: Vigilo AI Watch public record · last updated Oct 7, 2026
- Rogue behavior reported · on record↓falling
Confirmed entries carrying a rogue type, R1 to R4. The same count as the Rogue AI tracker.
Sources: Vigilo AI Watch public record; AI Incident Database · last updated Oct 9, 2026
- AI-related deaths or injuries reported · this year→steady
Verified entries in the physical-safety category: incidents where people were hurt or killed.
Sources: Vigilo AI Watch public record; AI Incident Database · last updated Aug 19, 2026
- 1 counter awaiting data
- AI scams (not counted yet)
Power
- AI regulations passed reported · this year↑rising
Verified entries classified as a law or binding rule on AI adopted.
Sources: Vigilo AI Watch public record · last updated Oct 20, 2026
- 3 counters awaiting data
- Governments deploying AI in military or surveillance (not counted yet)
- AI regulations stalled or struck (not counted yet)
- AI lawsuits filed (not counted yet)
People
- Weekly ChatGPT users est. · now↑rising
Estimate, as of Sep 19, 2026
Straight line between the two most recent weekly-user figures OpenAI has published, shown one interval behind real time. OpenAI gives each as "more than", so the line runs between floors. No chatbot maker publishes daily users.
- 900,000,000, Mar 31, 2026, OpenAI, Accelerating the next phase of AI
- 1,000,000,000, Aug 31, 2026, OpenAI, A milestone in expanding access to AI
- 1,200,000,000, Oct 7, 2026, OpenAI, GPT-6 and Intelligent UI for everyone
Sources: OpenAI · last updated Oct 7, 2026
- Weekly users, all assistants (published) est. · now↑rising
Estimate, as of Oct 29, 2025
The sum of the latest user figure each AI assistant's maker has published itself, weekly where the maker gives weekly, monthly where it gives only monthly. Every maker's part moves along the line between its own two newest published figures, one interval behind real time, as Weekly ChatGPT users does. A maker with one figure holds at it. Makers that publish no figure are listed and left out. People who use more than one assistant are counted once for each.
- 1,104,792,014 weekly, Sep 19, 2026, OpenAI, ChatGPT (moving)
- 1,000,000,000 monthly, Aug 11, 2026, Google, Gemini app
- 1,000,000,000 monthly, Oct 29, 2025, Meta, Meta AI
- 150,000,000 monthly, Oct 29, 2025, Microsoft, Copilot family
- 117,000,000 monthly, Mar 31, 2026, xAI, Grok
- no published figure, Anthropic, Claude
- no published figure, Perplexity, Perplexity
- no published figure, DeepSeek, DeepSeek
Sources: OpenAI; Alphabet, Q2 2026 earnings remarks; Google, Gemini app at 1 billion monthly users; Meta, Q3 2025 earnings call; Microsoft, FY26 Q1 earnings call; Space Exploration Technologies, Form S-1; Space Exploration Technologies, Form 424B4 · last updated Sep 19, 2026
- US children 9 to 12 who have used an AI chatbot est. · latest figure→steady
Estimate
The share of US children aged 9 to 12 who have used an AI chatbot, from the latest Common Sense Media census, held until the next one.
- 58%, Mar 26, 2026, Common Sense Media Census: AI Use by Tweens and Teens, 2026
Sources: Common Sense Media Census: AI Use by Tweens and Teens, 2026 · last updated Mar 26, 2026
- Fraud complaints from people 60 and over reported · latest figure→steady
Complaints of internet-enabled fraud filed with the FBI by people aged 60 and over in the latest year reported, held until the next annual report. Complaints, not a count of victims.
- 201,266, Dec 31, 2025, FBI Internet Crime Complaint Center, 2025 IC3 Annual Report
Sources: FBI Internet Crime Complaint Center, 2025 IC3 Annual Report · last updated Dec 31, 2025
- Bot share of web traffic est. · now↑rising
Estimate, as of Oct 26, 2025
Straight line between Imperva's two most recent yearly measurements of the share of all web traffic that is automated, shown one interval behind real time. Web traffic, not social media alone.
- 51%, Dec 31, 2024, Imperva (Thales), Bad Bot Report 2026
- 53%, Dec 31, 2025, Imperva (Thales), Bad Bot Report 2026
Sources: Imperva (Thales), Bad Bot Report 2026 · last updated Dec 31, 2025
Full feed · snapshot 45, taken Oct 26, 2026