Short answer
What do these mean?
Strict: the firm admitted an outage of 2 hours or more on a core service, or a global outage. Or it confirmed a security incident that exposed customer data or stopped a service.
Loose: this also counts partial, regional, or single-feature outages of 2 hours or more. It also counts any confirmed security incident, including one at a vendor.
Confirmed only: an incident counts only if the firm's own status page, postmortem, or statement backs it. Add reported: credible press also counts.
Never counted: user complaints the firm did not confirm, flaws in products customers run themselves, and hacker claims nobody has verified.
The two groups overlap, but they are not the same
Pick a view. Click a firm to see its sources.
Dot color is live status from public feeds. Gray means no feed we can read. Drag to pan. Use the buttons, or click the map and scroll, to zoom.
If the claim were true, the left and right zones would be nearly empty. They are not.
Early access only
Both
Incident, no public access
Neither (comparison firms)
25 early access firms
25 comparison firms
Major incident None found Pending (not yet confirmed)
Big firms have more outages, with or without Mythos
Anthropic chose the biggest software, cloud, and bank firms on purpose. Big firms run more services. More services means more chances to break. Their outages also get more press.
So some overlap was always likely. The real test is a fair comparison. We picked 25 similar firms with no public Mythos access. We used the same search and the same rules.
Access is CONFIRMED for 20 firms and REPORTED for 5. The rates are our own counts from public sources.
Big incidents happened both before and after access
Each row is one early access firm. The diamond marks when its access began or was announced. Hollow dots came before access. Solid dots came after.
Access began Incident before access Incident after access
After access does not mean caused by access. No firm blamed Mythos or AI. Stated causes include power, cooling, bad settings, and software bugs. A few firms gave no cause.
Every firm, with sources
| Sources |
|---|
CONFIRMED means the firm or an official record said it. REPORTED means credible press or a tracker said it. SPECULATION means a claim with no evidence. UNVERIFIED means a claim nobody has confirmed.
Live status of the public feeds
We check these public status feeds from our server. A green dot means the firm reports no problem right now. This is the firm's own view, not ours.
Other status pages we could not read automatically
What we could not verify
- About 200 groups joined Glasswing. Only about 25 names are public. Some comparison firms may have secret access.
- Apple had a widespread outage on Nov 17. By about 9:54 AM PT, Apple's status page listed an iCloud sign-in issue. We still mark it pending.
- Citi and Bank of America problems on Nov 17 come only from user complaint trackers. We did not count them.
- We did not count items about Goldman Sachs or Adobe. Each had only one weak source.
- The reports that four banks tested Mythos rely on unnamed sources.
- We could not confirm when Mozilla got access.
- Some outage lengths come from sites that copy status pages, not from the firms.
- Microsoft's final incident report is due later this week.
Short report
Project Glasswing is real. Anthropic launched it on April 7, 2026. It gave early Mythos access to 11 named partners and about 40 other groups.
We built two lists. List A has 25 firms with public early access. The comparison list has 25 similar firms without it.
We now count an incident as CONFIRMED only when the firm's own page backs it. Press-only rows are REPORTED.
Confirmed only, strict rule: 4 of 25 early access firms had a major incident. So did 4 of 25 comparison firms.
Confirmed only, loose rule: 7 of 25 vs 4 of 25.
With reported sources added, strict gives 32% vs 48%. Loose gives 68% vs 56%.
The gap is small under every setting. It points different ways under different rules.
With reported sources, eight firms are in both groups under the strict rule. They are Microsoft, Amazon, Google, JPMorganChase, Cloudflare, Fortinet, IBM, and the Linux Foundation.
None of those firms blamed Mythos or AI. The idea that Mythos caused the outages is SPECULATION. We found no evidence for it.
Plain English
- Big firms got Mythos early. Big firms also break more often.
- Firms without early access had about as many big problems.
- No firm blamed Mythos or AI for an outage.
- The "circle" post is a hunch, not a finding.
Action steps to review
- Check Apple's status page again before sharing.
- Look for official statements from Citi and Bank of America.
- Read Microsoft's final incident report when it comes out.
- Check the Palo Alto, Trend Micro, and Zscaler rows against their own status pages.
- Pull fresh numbers on the viral post before quoting them.
- Find official firm links for the 26 rows that now show REPORTED.