AI safety company founded by ex-OpenAI researchers Dario and Daniela Amodei. Builds the Claude family of large language models with a focus on steerability, interpretability, and reducing harmful outputs.
Method: a rolling quarter (13 weeks ≈ 91 days). Each cell is one 7-day slice of Statuspage-standard uptime (1 − (critical + 0.3 × major) / window; minors and maintenance excluded; red when the slice dips below the stated SLA).
Claude — Incidents by Day last 30d
MinorMajorCritical
Total 7 incidents
0
1
2
3
4
06-2306-2707-0107-0507-0907-1307-1707-21
Worst Incidents
Ranked by severity-weighted duration (impact × wall-clock). Click any incident for the full update timeline.
Type of Failure
Categorised from the public incident headlines (outage / latency / auth / etc.).
Elevated errors
6
86%
Other
1
14%
When It Breaks
Incident starts by weekday × hour, UTC. Hover a cell for incident detail.
S
0
M
Monday07:00–08:00 UTC
2incidents
2 minor
Fable 5 requiring usage credits on Max plansJul 20
Elevated error rates for Opus 4.5Jul 20
Monday13:00–14:00 UTC
1incident
1 minor
Elevated errors on Haiku 4.5Jul 20
Monday14:00–15:00 UTC
1incident
1 minor
Elevated errors for Claude Opus 4.8Jul 20
4
T
Tuesday07:00–08:00 UTC
1incident
1 minor
Elevated errors on Haiku 4.5Jul 21
Tuesday10:00–11:00 UTC
1incident
1 minor
Elevated errors across modelsJul 21
Tuesday11:00–12:00 UTC
1incident
1 minor
Elevated errors across Fable 5Jul 21
3
W
0
T
0
F
0
S
0
00
06
12
18
HOUR OF DAY · UTC7 incident starts · hover any cell for detail
Hottest hour
Monday 07:00 2
Worst day
Monday 4
Worst time of day
07:00–08:00 3
How fast they fix things
Distribution of resolution times alongside the 12-week trend so you can see whether they're getting faster or slower.
Typical fix
1h 41m
half of incidents resolve faster
On a bad day
8h
9 in 10 resolve faster than this
Distribution
<15m
0
15m–1h
0
1–4h
4
4–24h
2
24h+
0
+1 ongoing (not counted above)
MTTR Trend (12 weeks)
Are they recovering faster or slower over time? Lower is better.