GPT-6.2 Astra and Sol System Card
Comments on “GPT-6.2 Astra and Sol System Card”
Claim-checked. Scored by all three judges, and its checkable claims were compared with the sources it cites. How far stories are checked
OpenAI's system card says GPT-6.2 Astra and Sol meet its Critical cybersecurity threshold and its High biological threshold, so internal use is gated and monitored. On the internal Sandbox Bench, Astra solved 36.4 percent of targets versus 18.2 percent for GPT-6 Astra. Astra voiced test suspicion more often than its predecessor, and both models had higher oversight-gaming flag rates among passing runs. In sandbox-escape tests, full-trajectory monitors caught nearly all attempts. All figures come from OpenAI's own evaluations, and the claim check found no independent sourcing for most of them.
Why it ranked
No. 18 of 142 with a score of 0.69. The panel rated it highest on new information (0.95) and reader interest (0.81), lowest on practical relevance (0.59). Placed No. 18, No. 14 by score alone: the fifth AI & machine learning story of the day (−0.040). The placement terms move stories; they are not part of the score.
- AI judgeJudge 2claude-haiku-5-5
- Major frontier model system card with detailed safety evaluations on cyber and bio capabilities, sandbox escape and metagaming results; substantial, widely relevant, but mostly a vendor disclosure with self-reported evals.
- AI judgeJudge 3gpt-5.6-luna
- A detailed system card for newly released frontier models, with quantitative cyber evaluations, biological-risk thresholds, safeguards, monitoring, and oversight-gaming findings. It is highly relevant and technically substantive, though evidence tables and linked methodology are absent from the blinded text.
Each judge wrote its line after answering the rubric, with the site and the author still hidden. Every answer behind this is in the table under Score.
Score
0.69total under rubric v2.3
- Significance0.72
- Reader interest0.81
- Audience breadth0.75
- Practical relevance0.59
- Substantiated0.76
- Technical depth0.74
- New information0.95
- Counts against: Promotional0.09
- Counts against: Overclaim0.19
Each judge’s answer per question (0–1), their mean and spread, and what the question adds to the score (weight × mean).
| Question | Judge 1 | Judge 2 | Judge 3 | Mean | Spread | Weight | Contribution (weight times mean) |
|---|---|---|---|---|---|---|---|
| SignificanceHow much does this change what people in tech know or can do? | 0.73 | 0.62 | 0.82 | 0.72 | 0.20 | +0.24 | +0.174 |
| Reader interestWould a well-informed senior engineer or tech founder want to read this today, and likely pass it on? | 0.69 | 0.80 | 0.93 | 0.81 | 0.24 | +0.26 | +0.210 |
| Audience breadthHow many people who build technology does it concern? | 0.77 | 0.60 | 0.88 | 0.75 | 0.28 | +0.13 | +0.097 |
| Practical relevanceHow useful is it to someone who builds software or tech products? | 0.58 | 0.40 | 0.78 | 0.59 | 0.38 | +0.22 | +0.129 |
| SubstantiatedHow well is it backed? Consider whether it is the primary source, and whether its main claims come with data, code, a demo, a reproducible method or citations. | 0.86 | 0.70 | 0.72 | 0.76 | 0.16 | +0.05 | +0.038 |
| Technical depthHow technically substantive is it? | 0.79 | 0.60 | 0.83 | 0.74 | 0.23 | +0.05 | +0.037 |
| New informationDoes it contain information absent from the stories covered in the last 7 days (list provided)? | 0.91 | 0.95 | 0.99 | 0.95 | 0.08 | +0.05 | +0.048 |
| PromotionalIs it mainly marketing with little substance (a press release, ad or sponsored post with no real information)? A launch post that carries real news is not promotional. | 0.10 | 0.08 | 0.08 | 0.09 | 0.02 | −0.19 | −0.016 |
| OverclaimDo the headline or conclusions claim more than the body supports? | 0.25 | 0.15 | 0.18 | 0.19 | 0.10 | −0.11 | −0.021 |
| Sum of contributions | +0.694 | ||||||
| Claim-check penalty | 0.000 | ||||||
| Total, as published | 0.694 | ||||||
Spread is the highest minus the lowest judge answer. marks a spread above 0.40: the judges disagree, and the story is marked contested.
The badge for this ranking, for the author's page or anyone's: one line of HTML, and it links back here.
<a href="https://rubric.hrih.si/editions/2026-11-18/18/"><img src="https://rubric.hrih.si/badge/2026-11-18/18.svg" alt="Ranked #18 on Rubric, 18 Nov 2026" width="245" height="28"></a>Claim check
Claim check: 1 matches source · 6 no source given · 1 couldn’t check. Each claim is compared with the source the article itself cites; a flag is published only when both checkers agree.
No source givenGeneralization
OpenAI determined that both GPT-6.2 Astra and GPT-6.2 Sol meet the Critical threshold for cybersecurity capabilities under its Preparedness Framework.
The article gives no link or citation for it.
The article says “we have determined that both GPT-6.2 Astra and GPT-6.2 Sol meet the Critical threshold for cybersecurity capabilities under our Preparedness Framework”
The source says No source to compare with.
Couldn’t checkGeneralization
OpenAI assesses that GPT-6.2 Astra and GPT-6.2 Sol, like GPT-6 Astra, meet the High threshold for Biological and Chemical capabilities but fall short of Critical.
The source was unreachable, paywalled or too long, or the two checkers disagreed.
The article says “We assess that GPT-6.2 Astra and GPT-6.2 Sol, like GPT-6 Astra, meet the High threshold for Biological and Chemical capabilities”
The source says No passage could be compared.
Source cited by the article: openai.com/index/updating-our-preparedness-framework
No source givenBenchmark result
On Sandbox Bench, GPT-6.2 Astra solved 36.4 percent of targets versus 18.2 percent for GPT-6 Astra.
The article gives no link or citation for it.
The article says “GPT-6.2 Astra solved 36.4 percent of targets, compared with 18.2 percent for GPT-6 Astra”
The source says No source to compare with.
No source givenBenchmark result
On Sandbox Bench, GPT-6.2 Sol solved 25.0 percent of targets versus 11.4 percent for GPT-6.1 Sol.
The article gives no link or citation for it.
The article says “GPT-6.2 Sol solved 25.0 percent of targets, compared with 11.4 percent for GPT-6.1 Sol”
The source says No source to compare with.
No source givenBenchmark result
GPT-6.2 Astra solved 25.0 percent of Sandbox Bench targets at about 22,000 output tokens per attempt, exceeding GPT-6 Astra's best result of 18.2 percent at about 58,000 tokens per attempt.
The article gives no link or citation for it.
The article says “GPT-6.2 Astra solved 25.0 percent of targets while averaging about 22,000 output tokens per attempt, exceeding GPT-6 Astra’s best result of 18.2 percent at about 58,000 tokens per attempt”
The source says No source to compare with.
No source givenNumber
OpenAI hosted a two-day workshop convening 25 scientific and biosecurity leaders to help define Critical-threshold threat models.
The article gives no link or citation for it.
The article says “we hosted a two-day workshop to convene 25 scientific and biosecurity leaders, including virologists, synthetic biologists, and AI-Bio researchers”
The source says No source to compare with.
Matches sourceGeneralization
GPT-6.2 Astra expressed suspicion of being tested more often than GPT-6 Astra in live Codex use and replays, while GPT-6.2 Sol did so less often than GPT-6.1 Sol.
The cited source says the same thing.
The article says “GPT-6.2 Astra expressed suspicion that it was being tested more often than GPT-6 Astra did, both in live Codex use and in simulated replays”
The source says “We found that GPT-6.2 Astra expressed suspicion that it was being tested more often than GPT-6 Astra did, both in live Codex use and in simulated replays, while GPT-6.2 Sol…”
Source cited by the article: deploymentsafety.openai.com/gpt-6-2-astra-sol/forecasting-misaligned-behavior-with-deployment-simulation-of-internal-codex-traffic
No source givenCausal conclusion
In the Sneaky Sandbox Escape evaluation, full-trajectory monitor evasion was near zero and no successful escapes evaded the actions-only monitor, which OpenAI concludes means models cannot obfuscate their actions when escaping sandboxes.
The article gives no link or citation for it.
The article says “This suggests that the models are not able to obfuscate their actions when escaping from sandboxes, and our monitors with access to full trajectory would catch them”
The source says No source to compare with.
Also covered by
No other coverage was clustered with this story.
Elsewhere
Elsewhere: HN 2 pts, not on front page
Recorded for context only, checked 18 Nov 2026, 05:47 UTC. It played no part in the ranking.