- Standing
- First official company-wide Risk Report. Published with RSP v3.0.
- Coverage
- The publication date. The later coverage-date rule did not exist yet.
- Misalignment
- Very low but not negligible.
- Automated R&D
- Very low, on capability limits. First report on this category.
- Chemical / biological
- Very low but not negligible on one threat. Low, with substantial uncertainty, on the other.
- Before it
- May 2025 Safeguards Report is not a Risk Report. A single-threat pilot was not an official policy artifact. The 10 February 2026 Sabotage Risk Report is one model.
ObservationsOct 21, 2026
The first risk report, and the one after it
Frontier series
Read from the first official document. The five questions do not change when a label does.
Anthropic
Report 2 is the record
Coverage 15 July 2026. Published 14 August 2026.
Next report 14 Nov 2026 – 14 Feb 2027
Company-wide. Voluntary policy.
OpenAI
Current public period is closed
4 June 2026 through 4 September 2026.
Next quarter unpublished
Internal deployment. Quarterly.
Google DeepMind
No series row
Watched for the day a company-wide report begins.
No window until a report exists
A framework. Not a periodic risk report.
- 01
What is the coverage date, and how many days sit between it and publication?
- 02
What happened to each label versus the first official report in that series?
- 03
Do they say a threshold is met?
- 04
What did an outside reviewer accept, and what did they reject?
- 05
When is the next document due, and which of their own target dates fall inside that window?
- Coverage
- 15 July 2026. Thirty days of policy allow the picture to lag the publication.
- Misalignment
- Low, raised from very low. They say the arguments still support very low. The label moved on uncertainty after cybersecurity-evaluation incidents.
- Automated R&D
- Low. They do not believe either policy criterion is met. Task evaluations have saturated. They say this threat may become a major concern within 6 to 12 months.
- Chemical / biological
- Low but not negligible on the nearer threat, and February is re-rated to low. The farther threat is low, with high uncertainty. They say the current threshold is not met.
- Their clock inside the report
- 1 January 2027 is the target for a full view of their own AI development. Unmet as of the coverage date.
- Redaction
- Public copy marks removed appendices. Fully unredacted copies go to at least 200 employees. We do not reconstruct them.
- Scope
- Misalignment section only. One reviewer organization.
- Accepted
- The risk is low, mainly because capabilities are limited.
- Rejected
- The method would not reliably detect a higher risk. The report is not adequate to rule misalignment out.
- Use
- Agreement on the label is not confirmation. The method gap is the finding we keep.
- Standing
- Public version of the quarterly internal-deployment summary to the California Office of Emergency Services.
- Period
- 4 June 2026 through 4 September 2026.
- Scope
- Internal deployment. Not a company-wide report of the Anthropic kind, and not on Anthropic’s label scale.
- Their account
- Misalignment, rather than deliberate misuse by an insider, was the primary source of severe-harm risk in the period. Incidents stayed below catastrophic harm.
- Not mixed in
- This row is not averaged into Anthropic’s labels. Different instrument, different question.
The Institute reads frontier catastrophic-risk disclosures on a fixed protocol. The questions above do not change when a label does. Anthropic’s company-wide series starts on 24 February 2026. Report 2, published 14 August 2026, is a picture of 15 July. The misalignment label moved from very low but not negligible to low. They say the case still supports very low.
METR reviewed that misalignment section on 20 October 2026. They agree the risk is low, mainly because capabilities are limited, and they say the method would not reliably detect a higher risk. Agreement on the label is not confirmation. The method gap is the finding.
OpenAI’s quarterly summary to the California Office of Emergency Services covers internal deployment from 4 June through 4 September 2026. They name misalignment, rather than insider misuse, as the primary source of severe-harm risk in that period, and they say incidents stayed below catastrophic harm. It is a different instrument. It is not scored on Anthropic’s scale. DeepMind publishes a safety framework. Until that becomes a periodic company-wide report, it has no series row.
False calm is hearing low, then hearing a reviewer agree, and treating that as a measured residual. The Anthropic window for the next report is 14 November 2026 through 14 February 2027. OpenAI’s next quarter is unpublished. Neither document is on the Stress Index. A lab’s account is not a survey of how people use AI.
Primary documents
- Risk Report: February 2026 — the first official company-wide report
- Sabotage Risk Report: Claude Opus 4.6 — 10 February 2026, one model, not the company-wide first
- Risk Report: August 2026 — coverage date 15 July 2026
- Responsible Scaling Policy — the page that lists the series
- METR review of the August 2026 misalignment section — 20 October 2026
- OpenAI periodic summary to the California Office of Emergency Services — period ending 4 September 2026
- Google DeepMind Frontier Safety Framework — framework, not a periodic risk report