# METR asks whether discoveries are accelerating [Canonical reading](https://escapemeridian.com/readings/metr-asks-whether-discoveries-are-accelerating-5deff3d6) METR’s research update, published August 14, 2026, examines whether public records show faster discovery. The researchers report a sharp rise in disclosed vulnerabilities, possible acceleration in mathematics, and no clear acceleration in algorithmic optimization. They define “discovery” broadly to include advances in public knowledge, such as more efficient algorithms. The update compares time series across these areas, looking for changes in their rates. It says the vulnerability increase is not matched by a similar rise in known exploited vulnerabilities. For mathematics, the authors call evidence from solved open problems weak and note that historical baselines are difficult to establish. For optimization, they find no appreciable acceleration in the public record, despite examples of AI contributions. The authors caution that the analysis uses public discoveries, that data collection and analysis were performed by agents, and that mistakes likely remain. They also say labs may be making discoveries they do not disclose. Whether the observed differences reflect AI’s effectiveness, spending, disclosure practices, or data quality remains unresolved. Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims. ## Original source METR · Article Original source date: Aug 14, 2026 (2026-08-14T07:00:00.000Z). Date basis: Date supplied by the publisher’s feed. Added to Escape Meridian: 2026-10-29T17:28:01.451Z. First observed: 2026-10-18T02:43:52.712Z. Retrieved: 2026-10-24T01:25:39.035Z. - [Have We Seen an Acceleration in Discoveries?](https://metr.org/notes/2026-08-14-llm-contribution-to-discoveries/) ## Limits Text as retrieved on October 24, 2026; originally published August 14, 2026. The researchers say the analysis and data collection were performed by agents, mistakes likely remain, and the data series are difficult to interpret. The analysis focuses primarily on public discoveries; the extent of undisclosed progress inside AI labs is unknown. The source is an unpolished research update and says it has gone through less review than METR’s research posts. ## Open question How much of the difference between domains reflects AI’s contribution, rather than disclosure practices or gaps in the historical data? ## Source-stated METR reports a sharp rise in public vulnerability disclosures, while growth in databases of exploited vulnerabilities is lower. > The rate of vulnerabilities reported across many projects has dramatically accelerated in 2026 compared with 2025 - [Supporting source](https://metr.org/notes/2026-08-14-llm-contribution-to-discoveries/) ## Source-stated METR describes the evidence for faster mathematical discovery as weak and says reliable historical baselines are difficult to construct. > This is some evidence of acceleration, but it is weak. - [Supporting source](https://metr.org/notes/2026-08-14-llm-contribution-to-discoveries/) ## Source-stated METR reports no appreciable acceleration in the public record of overall algorithmic optimization. > Thus the public record of overall algorithmic optimization does not yet show an appreciable acceleration. - [Supporting source](https://metr.org/notes/2026-08-14-llm-contribution-to-discoveries/) ## Source-stated The researchers say undisclosed discoveries inside AI labs could make public records understate progress. > It’s possible that the rate of discoveries is accelerating more rapidly than we observe in public data. - [Supporting source](https://metr.org/notes/2026-08-14-llm-contribution-to-discoveries/) ## Source and contributor context METR · Publisher The publisher is identified here. No individual author has been inferred. ## Source media Media retrieval status: unchecked. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Time-series Observations arranged over time, such as monthly sales; forecasting uses their patterns to estimate later values. - [Definition reference](https://developers.google.com/machine-learning/glossary#time-series-analysis) ## METR METR studies what AI systems can do and how reliably they complete tasks, including work with human developers. - [Explore this context](/entities/metr) - [Reference](https://metr.org/) ## Editorial provenance Version 1. Publication ID: fcaa83dd-8a50-3ec9-a86c-e5854449c7a2. Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5. Source check: passed by anthropic/claude-opus-5.5 on 2026-10-29T17:27:48.866Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: bridge\_admitted\_by\_model, second\_opinion\_categorical. Observation: 0d4964e71905817c46600acb92a8c18f6e687f73ec01580530cb2cf899f1ecc4. Source text hash: 9914324bd61c791a706fda32d9e52c3bc7fbcdcf9e3f5b326b50a9fe3b4a8816. ## Revision and source-check history ### Version 1 — 2026-10-29T17:28:01.451Z METR’s research update, published August 14, 2026, examines whether public records show faster discovery. The researchers report a sharp rise in disclosed vulnerabilities, possible acceleration in mathematics, and no clear acceleration in algorithmic optimization. They define “discovery” broadly to include advances in public knowledge, such as more efficient algorithms. The update compares time series across these areas, looking for changes in their rates. It says the vulnerability increase is not matched by a similar rise in known exploited vulnerabilities. For mathematics, the authors call evidence from solved open problems weak and note that historical baselines are difficult to establish. For optimization, they find no appreciable acceleration in the public record, despite examples of AI contributions. The authors caution that the analysis uses public discoveries, that data collection and analysis were performed by agents, and that mistakes likely remain. They also say labs may be making discoveries they do not disclose. Whether the observed differences reflect AI’s effectiveness, spending, disclosure practices, or data quality remains unresolved. ## Original source METR · Article Original source date: Aug 14, 2026 (2026-08-14T07:00:00.000Z). Date basis: Date supplied by the publisher’s feed. Added to Escape Meridian: 2026-10-29T17:28:01.451Z. First observed: 2026-10-18T02:43:52.712Z. Retrieved: 2026-10-24T01:25:39.035Z. - [Have We Seen an Acceleration in Discoveries?](https://metr.org/notes/2026-08-14-llm-contribution-to-discoveries/) ## Limits Text as retrieved on October 24, 2026; originally published August 14, 2026. The researchers say the analysis and data collection were performed by agents, mistakes likely remain, and the data series are difficult to interpret. The analysis focuses primarily on public discoveries; the extent of undisclosed progress inside AI labs is unknown. The source is an unpolished research update and says it has gone through less review than METR’s research posts. ## Open question How much of the difference between domains reflects AI’s contribution, rather than disclosure practices or gaps in the historical data? ## Source-stated METR reports a sharp rise in public vulnerability disclosures, while growth in databases of exploited vulnerabilities is lower. > The rate of vulnerabilities reported across many projects has dramatically accelerated in 2026 compared with 2025 - [Supporting source](https://metr.org/notes/2026-08-14-llm-contribution-to-discoveries/) ## Source-stated METR describes the evidence for faster mathematical discovery as weak and says reliable historical baselines are difficult to construct. > This is some evidence of acceleration, but it is weak. - [Supporting source](https://metr.org/notes/2026-08-14-llm-contribution-to-discoveries/) ## Source-stated METR reports no appreciable acceleration in the public record of overall algorithmic optimization. > Thus the public record of overall algorithmic optimization does not yet show an appreciable acceleration. - [Supporting source](https://metr.org/notes/2026-08-14-llm-contribution-to-discoveries/) ## Source-stated The researchers say undisclosed discoveries inside AI labs could make public records understate progress. > It’s possible that the rate of discoveries is accelerating more rapidly than we observe in public data. - [Supporting source](https://metr.org/notes/2026-08-14-llm-contribution-to-discoveries/) ## Source and contributor context METR · Publisher The publisher is identified here. No individual author has been inferred. ## Source media Media retrieval status: unchecked. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Time-series Observations arranged over time, such as monthly sales; forecasting uses their patterns to estimate later values. - [Definition reference](https://developers.google.com/machine-learning/glossary#time-series-analysis) ## METR METR studies what AI systems can do and how reliably they complete tasks, including work with human developers. - [Explore this context](/entities/metr) - [Reference](https://metr.org/) ## Editorial provenance Version 1. Publication ID: fcaa83dd-8a50-3ec9-a86c-e5854449c7a2. Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5. Source check: passed by anthropic/claude-opus-5.5 on 2026-10-29T17:27:48.866Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: bridge\_admitted\_by\_model, second\_opinion\_categorical. Observation: 0d4964e71905817c46600acb92a8c18f6e687f73ec01580530cb2cf899f1ecc4. Source text hash: 9914324bd61c791a706fda32d9e52c3bc7fbcdcf9e3f5b326b50a9fe3b4a8816. ### passed — 2026-10-29T17:27:48.866Z — anthropic/claude-opus-5.5 Publication: fcaa83dd-8a50-3ec9-a86c-e5854449c7a2 attribution — pass: The reading attributes the work to METR researchers and frames the findings as their reported observations. The source is a METR Notes post, and the draft's limitations disclose its lighter review. The source's caveat that notes may not reflect METR as a whole is not repeated, but the attribution is not misleading. > Notes are unpolished research updates directly from METR researchers, and have gone through less review than our research posts. summary — pass: The summary accurately conveys the three headline findings, the slope-change approach, the gap between disclosed and exploited vulnerabilities, the caveats about public data and agent-performed analysis, and the candidate explanations. Calling math acceleration 'possible' slightly softens the source's 'likely' but remains faithful. The title and question follow from the source. The text is understandable without specialist background. > Discovery of cyber vulnerabilities has accelerated sharply. date — pass: The draft gives August 14, 2026, which matches evidence.originalPublished (UTC day) and the date in the URL. The source text references data through early August 2026, which is consistent. The reading uses no hindsight framing. The evidence is a publisher page, not an archive copy, so no archive disclosure is required. > Microsoft-issued CVEs grew from 1,243 in all of 2025 to 1,927 through 2026-08-11 support — pass: Each claim is directly stated in the source with a matching excerpt. The lower growth in exploited-vulnerability databases, the weak math evidence, the lack of optimization acceleration and possible undisclosed lab progress are all source-stated. The unresolved-causes sentence and the question reflect the source's own list of candidate explanations. The reading adds no unsupported benefits or costs. > databases tracking exploited vulnerabilities (CISA and Vulncheck KEVs) show significantly lower year-over-year growth than the databases of known vulnerabilities media — pass: No media is selected, and no verified media candidates were supplied, so there is no media identity or relevance issue. ## Recorded references Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter. No public references recorded here yet. [Reference data (JSON)](/readings/5deff3d602149d146675a7f3253d4f36da620ce7b1e8ef7a9563f285a2c6dd33/references.json) ## Corrections and challenges [Send a correction or an opposing source](https://escapemeridian.com/corrections/) [JSON record](https://escapemeridian.com/readings/metr-asks-whether-discoveries-are-accelerating-5deff3d6.json)