A lab says AI has made its research 5× faster

Where things stand (mid-October 2026)

No covered developer has published a developer-wide, AI-attributable research-pace multiplier anywhere near 5, and the labs that measure this say they are well below 2.

• Anthropic is the only lab with a quantified, published method. Its August 2026 Risk Report (Aug 14; data through Jul 15) finds "meaningful acceleration starting in early-to-mid 2025, though by less than a factor of 2." It treats its survey result (about 4x self-reported uplift, geometric mean) as not translating into research progress. The Claude Opus 5.5 System Card (Sep 22) finds "no sustained AI-attributable 2× acceleration." Its capability-index (AECI) analysis gives a trend-break of 14.4 to 22.2 points/yr, which is 1.53x (95% range 1.20–1.82), but a one-time-jump model fits better in 99 of 100 refits. METR's preliminary estimate for Anthropic is about 1.5x overall ("1.5 years in 1 year"), with roughly a 30% chance of 2x.
• Google DeepMind's Gemini 4 Argon safety report (October 2026) finds "no significant end-to-end acceleration." Its capability-index slope went from 0.091 to 0.118 points/month, which is not statistically significant.
• OpenAI's Research acceleration post (Sep 6, before the window opened) reports only input metrics, such as 3.1 agent-workdays per human workday, and explicitly gives no overall multiplier. The GPT-6 Astra system card says Astra is below "High" for AI self-improvement. OpenAI's own "Critical" self-improvement threshold is exactly a 5x speed-up ("1/5th the wall-clock time" of 2024 progress), and crossing it means halting development.
• Other Anthropic publications (When AI builds itself, Measurements of pace) report output measures: 8x code per engineer, AI leading 26% of R&D work, none fully autonomous. Those are excluded by the question. On Oct 14, METR wrote that no company has published detailed evidence on overall acceleration.

How far 5x is (my own arithmetic)

At Anthropic's baseline of 14.75 points/yr, 5x means about 74 AECI points per year. In Anthropic's framework a doubling means two years of baseline progress in one year, so 5x means five years in one. Assume compute is held equal in the human-only comparison and grows 3–4x a year. Then a 5x overall ratio needs algorithmic progress about 9–10x faster than baseline. That is the middle of what Anthropic calls its "most acute" intelligence-explosion regime (10³–10¹⁰× effective scale-up per year). In practice, a YES requires an intelligence explosion that a lab measures, attributes to AI, and publishes.

Reference class and adjustments

The labs' own measured multiplier went from about 1.0 to about 1.5 between 2025 and 2026, even though inputs grew fast (merged code changes rose 12x from May to September 2026). That points to strong bottlenecks from compute and from the parts of research AI can't yet do.

Reasons it could happen:

• Insiders expect much faster progress. Jack Clark (Oct 2026) thinks full self-improvement is possible in 2027. OpenAI is targeting an automated researcher by March 2028. Dario Amodei says progress has been "drastically faster" since summer 2026.
• The AI Futures Project's April 2026 medians for an "automated coder" were mid-2028 (Kokotajlo) and mid-2030 (Lifland).
• Anthropic publishes risk reports every 3–6 months using a quantitative method, and OpenAI has committed to being transparent about its self-improvement progress.

Reasons it might not:

• Labs openly plan to slow down ("pacing") before this regime. OpenAI's framework halts development at exactly 5x.
• Measuring a no-AI counterfactual and attributing the gain to AI are hard, so labs tend to report conservative bounds such as "more than 2x."
• Material may be redacted or, in a crisis, government-controlled.
• Chinese labs rarely publish this kind of measurement.

Model

Scenario	Weight	Chance of a qualifying statement by 2031
Fast takeoff, 2027–28	~30%	~0.65 × 0.65 (not slowed down, then published)
Moderate, 2029–31	~35%	~0.25 (partly falls after the window)
Slow, plateau or strongly paced	~35%	~3%

Adding about 2% for generous-but-qualifying methods gives roughly 20–25% by end-2031. Quarterly hazards are about 0.2–0.8% through 2027 (the next risk reports start from below 2x). They peak at about 1.7% in late 2028 to 2029, when most fast-scenario reports would land, then ease to about 1.3%. Cumulative probability: 2.3% by end-2027, 7.3% by end-2028, 13.3% by end-2029, 22.4% by end-2031.

Strongest case against: This world looks like the early stage of an intelligence explosion, and 5+ years is long, so 35–45% is defensible. I don't go that high because the labs' own measured pace has barely moved, the labs say they will slow down, and the resolution bar is strict.

Question. By the last day of each quarter, has a covered developer stated, giving a number and a method, that AI systems have made the overall pace of its own AI research at least 5 times faster than it would be with human researchers alone?

Definitions.

• Overall pace of its own AI research: the developer-wide rate of substantive progress in developing or improving AI models and methods, measured per unit of calendar time. For a covered developer that is a named AI lab or model organization within a larger company, the scope is that organization's AI research, not the parent's unrelated work. An aggregate estimate from a sample counts only if the developer explicitly presents it as an estimate of its overall AI research pace.
• Human researchers alone: a counterfactual in which the same human research workforce has ordinary non-AI software and research infrastructure but no AI systems performing or assisting research work. This does not exclude training or evaluating AI models as objects of research. The comparison must distinguish AI assistance from changes in human staffing, training compute, or other non-AI resources; a historical baseline counts only if the method explains how those differences are accounted for.
• At least 5 times faster: a ratio of AI-assisted research progress per unit time to the human-only counterfactual of at least 5.0, or equivalent time to the same progress of at most one-fifth. The AI systems need not have been built by the developer making the statement.
• Giving a number and a method: the publication reports the multiplier (or quantities explicitly equated to it), identifies the research-progress measure and observation period, and explains how both the AI-assisted result and human-only counterfactual were estimated and aggregated to the developer-wide level. Experimental comparisons, observational analyses, and explicitly described estimation models may count; public raw data, independent replication, and statistical significance are not required. Survey evidence counts only when the developer uses a disclosed method to translate it into and endorse an overall research-progress estimate, rather than merely reporting respondents' output estimates.

Resolves YES if a qualifying developer-authored publication states that an achieved or currently observed overall multiplier is at least 5.0 and supplies the method above, directly or through linked developer-authored documentation. A stated central estimate of at least 5.0 counts despite an uncertainty interval extending below 5.0. If only a range is given, its lower bound must be at least 5.0. The Observation period convention applies.

Exclusions (resolve NO).

• Forecasts, targets, hypothetical scenarios, claims of potential acceleration, and model capabilities not translated into achieved research acceleration.
• Multipliers for coding speed, code volume, experiment counts, benchmark task performance, individual researchers, selected teams, or selected projects without the required developer-wide research-progress estimate.
• An unadjusted average or median of employees' self-reported output gains; raw output is not itself overall research progress.
• Faster model releases, improved model scores, or a before-and-after increase in research pace without an explicit AI-attributed comparison against the human-only counterfactual.
• Statements such as “up to 5x,” or a numerical claim accompanied only by an assertion that internal analysis was performed, without the specified method.
• In the absence of a qualifying statement by a horizon, that horizon resolves NO.

Resolving source. Covered-developer-authored publications only: official website publications, research papers, system cards, risk reports, or public regulatory filings. Public confirmation (A2) applies with this narrower author requirement; third-party estimates do not independently qualify. Apply the Resolution window and Resolution date (A3) conventions. If the claim and required method are published separately, use the publication date on which both first become publicly available in qualifying sources.