Offline Brain / AI Watch
How often the AI and cloud services people rely on go down, read from each provider's own status page. Every item is dated and linked. We don't guess, and we don't give it a risk score. Not an official alert service.
112 incidents in 30 days across four services millions of people rely on. Each one links to the provider's own post.
1. AI and cloud incidents, last 30 days
112
across OpenAI, Claude, Cloudflare and GitHub, counted from their own incident histories.
An incident is anything the provider chose to post, from a short slowdown to a full outage. We count them; we don't score them.
2. AI in the news
Each status is the provider's own wording, copied from its status page at the time shown. Click through for theirs.
All Systems Operational
Latest: Elevated errors with search in Dots, ChatGPT Work, and certain ChatGPT thinking experiences 31 Oct
Source: status.openai.com · read 15:08 NZT ↗All Systems Operational
Latest: Elevated errors affecting the Google Calendar connector 29 Oct
Source: status.claude.com · read 15:08 NZT ↗Partial System Outage
Latest: Cache purge delays in Qingdao Shandong, China 31 Oct
Source: www.cloudflarestatus.com · read 15:08 NZT ↗All Systems Operational
Latest: Incident with Dependabot 31 Oct
Source: www.githubstatus.com · read 15:08 NZT ↗No open incidents
Latest: Multiple services were experiencing issues in the europe-west2 region. 20 Oct
Source: status.cloud.google.com · read 15:08 NZT ↗No posts in the last 24 h
Latest: Service disruption: Increased Error Rates 15 Sept
Source: health.aws.amazon.com · read 15:08 NZT ↗The newest entries in the AI Incident Database, a public record of real harms and near misses involving AI. We leave out cases involving death, self-harm or abuse.
Source: AI Incident Database · read 1 Nov, 15:08 NZT
Dated statements from lab leaders and researchers, each linked to where they said it. We quote them. We don't make predictions of our own.
Wrote that it is possible we will have superintelligence “in a few thousand days”
Wrote that powerful AI, smarter than a Nobel winner in most fields, “could come as early as 2026” though it could also take much longer.
Said AI that matches or beats humans at any task will start to emerge in the next five to ten years.
Signed a one-line statement that mitigating “the risk of extinction from AI” should be a global priority, alongside pandemics and nuclear war.
Real events in AI since ChatGPT, each with its source. We add to it as things happen.
OpenAI's chatbot reaches the public. The AI race goes mainstream.
Source ↗OpenAI's next model passes exams that earlier AI couldn't.
Source ↗An open letter asks labs to pause training AI more powerful than GPT-4 for six months. They don't.
Source ↗The heads of the three leading AI labs sign: AI extinction risk should be a global priority.
Source ↗28 countries and the EU agree frontier AI could cause serious, even catastrophic, harm.
Source ↗The first broad law on AI comes into force in Europe.
Source ↗OpenAI's CEO writes that superintelligence is possible within a few thousand days.
Source ↗Geoffrey Hinton shares the Physics prize, and uses the stage to warn about AI.
Source ↗Anthropic's CEO writes that powerful AI could arrive as early as 2026.
Source ↗Over 90 experts led by Yoshua Bengio publish the first global report on advanced AI risks.
Source ↗Google DeepMind's CEO says human-level AI will start to emerge within five to ten years.
Source ↗Your own AI, Wikipedia, WikiMed and US mapsNZ mapsAustralian maps (being added), on your own laptop. It never checks a status page, because it doesn't need one.
Recent news headlines about harm from AI: deepfakes, scams, data leaks, bans, lawsuits and AI that got it wrong. Each is the headline as published, linked to the original. Read 31 Oct, 20:40 NZT.
Collected from the AI Incident Database and the public news feeds of The Register, Ars Technica, The Verge, BBC News and The Guardian, filtered to AI harms. Stories about deaths, self-harm or abuse are left out. Paywalled links are left out. The same story from two sources is shown once. A feed that can't be read is skipped.
A small job reads each source every 10 minutes and saves the result; this page shows the last read and its time. If a source can't be read, its tile says "feed unavailable", never zero. Sources: OpenAI, Anthropic, Cloudflare and GitHub status pages (Atlassian Statuspage), Google Cloud Service Health, the AWS Health feed and the AI Incident Database. Headlines come from the AI Incident Database and the public feeds of The Register, Ars Technica, The Verge, BBC News and The Guardian, kept as published. Names are used to identify services only. This is not an official alert service. In an emergency call your local emergency number (111 in New Zealand, 000 in Australia, 911 in the US); for civil defence updates in NZ see civildefence.govt.nz.