{ "question_id": "Q10", "slug": "bio-chem-weapons", "title": "Biological and chemical weapons", "type": "level", "levels": [ { "key": "B1", "label": "Top-tier developer threshold declared" }, { "key": "B2", "label": "AI-assisted acquisition or synthesis confirmed" }, { "key": "B3", "label": "Casualties" } ], "resolution": { "levels": [], "values": [], "annotations": [] }, "forecasts": [ { "question_id": "Q10", "forecast_date": "2026-09-14", "evidence_cutoff": "2026-09-14", "reading_adopted": null, "ambiguities": [ "The text calls the levels nested, although B2 can occur without an OpenAI or Anthropic declaration. I apply the explicit higher-implies-lower convention rather than requiring chronological B1 satisfaction.", "B2 specifies neither malicious intent nor a minimum quantity, and does not exclude authorized research. I include such research when all stated agent, materiality, and publisher requirements are met.", "The relative link containing the successor-definition rule was not supplied with its contents and could not be located. I adopt substantive equivalence rather than automatic acceptance of weaker successor labels.", "For B3, 'an agent for which B2 is satisfied' could be read as an agent type rather than the AI-enabled episode. I require linkage to that episode and exclude unrelated historical outbreaks.", "Anthropic's Fable 5.1/Mythos 5.1 system card (2026-09-01) formally applies CB-1 and states CB-2 was not crossed; CB-1 is excluded by the question, so B1 remains unresolved.", "Anthropic's September 2026 threat report documents actors using Claude in ways that 'could support biological weapons development' without asserting intent; whether such a report satisfies B2 depends on how strictly 'confirms uplift to an actual CB weapons effort' is read. I adopted the strict reading.", "If a cross-lab safety pact or framework revision restructures the top CB tiers, the successor-definition rule must be applied; I assume the closest successor to Critical/CB-2 is used.", "Successor definitions: B1 references Anthropic RSP v3.1 (2026-04-02) but says successor definitions follow the rule in the linked Anthropic AI R&D threshold question, which I cannot access. Anthropic has since revised the novel-CBW threshold (RSP v3.3, effective May 26, 2026) to 'functionally substituting for the scarce human expertise that is currently the primary barrier to novel development.' I assume the current/revised definition governs so long as it tracks the same underlying capability (novel catastrophic CBW production), and that definitional changes that eliminate the capability bar would not count. This lowers P(B1) relative to a fixed v3.1 reading.", "Anthropic's 'we act as though they meet our CB-X threshold' phrasing: I treat an unambiguous statement that Anthropic acts as though the novel-CBW (CB-2) threshold is met as counting as a public statement that the threshold is met, though it is more hedged than a direct declaration. Current such language applies only to CB-1 (excluded from B1).", "B2 materiality: I require the government body to state AI assistance was necessary to or substantially enabled the acquisition/synthesis; 'used' or 'consulted' does not count. Government charging documents rarely make such findings explicitly, which is a key reason B2 is low.", "B3 requires deaths attributed to 'a release' of a B2-qualifying agent; an intended-but-interdicted attack or a non-release laboratory exposure would not count.", "The referenced successor-definition rule in 'Anthropic's AI R&D threshold' is not reproduced in this question. Anthropic raised the Novel-CB bar in RSP v3.3 (2026-05-26) relative to the quoted v3.1 text; I treat the current named Novel-CB threshold as the successor rather than freezing v3.1 language. If v3.1 were frozen, Anthropic might never restate it, and B1 would then rest almost entirely on OpenAI Critical.", "B2 requires a government/court materiality finding ('necessary' or 'substantially enabled'), not mere use. That line is untested; many plausible incidents will be described as 'used AI' and resolve NO.", "B3 requires both B2 for the same agent and national-government attribution of ≥10 deaths. A classified or loosely worded attribution would not resolve.", "Anthropic revised the CB-2 threshold in RSP v3.3 (effective 2026-05-26) from the v3.1 wording quoted in the question ('significantly help... moderately resourced expert-backed teams... damages far beyond... COVID-19') to a harder, more specific bar ('functionally substitute for the scarce human expertise... would otherwise require recruiting one of a small number of world-leading specialists'). The question's successor-definition rule points to another question I cannot access. I adopted the reading that the operative CB-2 definition at the time of any statement governs and that a CB-2 'met' statement resolves B1 YES, since the v3.3/v3.4 text is comparable-or-stricter. If the intended rule freezes the v3.1 wording, my B1 curve would be modestly higher (Anthropic's own Frontier Safety Roadmap still uses the v3.1 wording under the heading 'Potential advances in the next few months').", "'Resolving sources. B1: developer-authored documents per A2' is ambiguous: A2's second bullet would allow an undisputed third-party evaluator or government AI safety institute report to constitute confirmation, but the phrase 'developer-authored documents' suggests only the first bullet. I adopted the broader reading, which adds roughly 2-3 points over five years.", "Whether a hedged-but-operative formulation such as Anthropic's CB-1 practice ('we currently act as though they meet our CB-1 threshold', 'provisionally meeting') counts as 'stating the threshold is met' for CB-2. I treat it as YES, since it is materially stronger than 'cannot rule out', which the question explicitly excludes. This is the single largest swing factor in my near-term B1 estimate.", "The question says rungs are cumulative and 'higher rungs imply lower', but B2's own text contains no dependency on B1 (only B3 explicitly depends on B2). If the resolver strictly requires B1 before B2, my B2 curve should be scaled by roughly P(B1 already confirmed), which would cut the 2027 values by half or more. I applied only a ~10% haircut and note the risk.", "B2's 'acquired, produced, or synthesized' is unclear as to whether a foiled plot in which the actor obtained precursor material or a small quantity but was interdicted before full synthesis qualifies, and whether an FBI/DOJ charging document (as opposed to an adjudicated finding) counts as a 'public confirmation'. I assumed a charging document or agency press release counts, but that the actor must have actually obtained/produced some quantity of the listed agent.", "OpenAI has been reported (Aug 2026) to be rewriting its Preparedness Framework. If the Critical biological-and-chemical category is renamed, restructured, or removed, an OpenAI-side B1 declaration may become impossible to map; I did not separately model this beyond a small downward adjustment in the 2029-2031 tail.", "Anthropic revised its 'novel chemical/biological weapons production' threshold in RSP v3.3 (2026-05-26) / v3.4 (2026-07-08); the question cites v3.1 (2026-04-02). I applied the question's stated rule that successor definitions govern.", "B2 is not logically dependent on B1, but the schema's nesting/monotonicity rule forces p(B2) <= p(B1); I adopted the schema reading (B1 set to be the max of its own value and B2/B3).", "B2's 'materiality' bar ('necessary to, or substantially enabled') is stringent and rarely stated by governments; I read it literally, so mere 'used/consulted AI' statements do NOT count." ], "key_drivers": [ "Latest major developer assessments fall short of the specified biological/chemical B1 threshold despite nearby lower-tier and cybersecurity milestones.", "Improving expert scientific assistance and laboratory automation create repeated opportunities for a qualifying developer declaration.", "Recent documented biological misuse increases the prospective investigation pipeline but does not itself satisfy B2.", "B2 can resolve through authorized government-reported research as well as criminal or state-program investigations.", "Government confirmation of actual synthesis and material AI assistance is a separate, often delayed or permanently absent step.", "Ten-death releases require additional dissemination, mortality, and attribution conditions; defenses and intervention can interrupt the chain.", "Successor-policy changes, national-security review, and a possible frontier-development slowdown affect publication timing.", "Anthropic formally applied CB-1 to Mythos 5.1 (2026-09-01) and explicitly found CB-2 not crossed, citing four failure modes — one rung short of B1.", "OpenAI affirmed Critical *cyber* for Astra (2026-09-01) one month after 'cannot rule out', showing labs will make affirmative top-tier designations; bio remains at High.", "Rapid capability trend: Terminal-Bench-Science doubled in one generation; Mythos 5.1 beats 75th-percentile ML-bio experts on RNA design; Anthropic Risk Report (Aug 2026) warns AI R&D acceleration is near.", "Bio thresholds are harder to evaluate and demonstrate than cyber; labs can honestly stay at 'cannot rule out' (which resolves NO) for extended periods.", "Anthropic's 2026-09-10 threat report is the first developer disclosure of real-world bio-adjacent misuse but stops short of asserting weapons intent — sets the template for a future B2-resolving disclosure.", "Post-incident environment (training pauses, Altman saying most advanced models cannot yet be safely deployed, cross-lab safety pact talks) could slow release cadence and delay resolving documents.", "OpenAI's GPT-6 Astra system card (Sep 3, 2026) declares Critical cyber while rating bio/chem only High — bio has lagged cyber by ~2 generations, and OpenAI has shown it will publicly declare a Critical threshold while continuing deployment, shortening B1 disclosure lag if bio crosses.", "Anthropic's Aug 2026 Risk Report and Fable/Mythos 5.1 system card explicitly state the novel-CBW (CB-2) threshold is NOT met (models meet only CB-1, which is excluded from B1); Anthropic publishes Risk Reports roughly every 6 months, so any crossing would surface quickly.", "Anthropic narrowed the CB-2 threshold in RSP v3.3 (May 2026) and uses 'act as though' framing rather than declarations — a threshold-gaming signal that reduces P(B1).", "SecureBio's VCT (June 2026) shows frontier models outperforming PhD virologists on wet-lab troubleshooting — the 'scarce expertise' barrier defining CB-2 is eroding, and Anthropic reports bio capability evals are saturating with early signs of AI-R&D acceleration.", "Rising documented bio-misuse attempts (Anthropic: 5 case studies in 8 months; Google: blocked synthesis-guide request) increase the B2 opportunity pool, while strong political attention on AI-enabled bio harm (FL AG investigation of OpenAI, congressional biosecurity push, German Bundestag report) raises the chance governments would state AI materiality explicitly in a confirmed case.", "No government has ever confirmed completed AI-assisted synthesis of a listed agent with a materiality finding, and no CBW attack in modern history has killed 10+ apart from rare exceptions (Aum sarin, 13) — the base rates anchoring B2 and B3 near zero.", "Neither OpenAI nor Anthropic has stated the novel/Critical bio-chem threshold is met as of 2026-09-14; both just reaffirmed High/CB-1 only (Astra 2026-09-03, Mythos 5.1 2026-09-01).", "OpenAI's PF v2 Critical bio bar ('enable an expert' to make a novel Class A-equivalent) is lower than Anthropic's current CB-2, and OpenAI has shown it will declare Critical (cyber, Astra).", "Anthropic already raised the Novel-CB threshold in May 2026; further goalpost-moving is the main reason B1 is not ~1 by 2031.", "B2 is gated by a government/court materiality finding, not by capability or by developer TI reports; chemical Schedule 1 is the more plausible listed-agent path than novel pandemic synthesis.", "B3 requires a ≥10-death release of a B2-qualified agent plus national attribution; historical CB attacks almost never clear that bar in a five-year window.", "B1: Anthropic's Aug 2026 Risk Report says models 'do not yet meet' the CB-2 (novel) threshold but 'may provide significant uplift' — one to two model generations from a plausible declaration; next risk report ~Feb 2027", "B1: OpenAI has rated bio/chem High since GPT-5 (Aug 2025) and remained High at GPT-6 Astra (Sep 2026), with zero of three novel-design evals clearing Critical at GPT-5.6; but OpenAI demonstrated willingness to declare Critical (cyber, Sep 2026), setting precedent", "B1: Both developers have incentives to delay/soften declarations (Anthropic revised its CB-2 threshold in v3.3/v3.4; declaring Critical triggers costly constraints), partially offset by the question's successor-definition rule and by mounting external scrutiny (June 2026 CEO letter to Congress, NYT/AISI uplift findings)", "B2: Attempts are already occurring and increasing (Anthropic's Sept 2026 threat report describes blocked bio misuse by actors including state-linked groups), but B2 requires actual acquisition/synthesis of a select agent/Schedule 1 chemical plus a government/court materiality finding — a compound, slow event", "B2/B3: Historical base rate of prosecuted select-agent incidents is a few per decade, and no court has yet made an 'AI substantially enabled' finding; government confirmation lags incidents by months to years", "B3: Requires B2 plus an actual release killing >=10 plus national-government attribution; no bioterror attack has killed 10+ in the modern era (2001 anthrax letters killed 5), and most plots are interdicted pre-release", "Physical and biological constraints on novel pathogen development and laboratory robotics vs pure in silico software iteration", "OpenAI and Anthropic frontier evaluations (e.g. GPT-6 Astra and Mythos 5.1 both reaching High/CB-1 but remaining explicitly below Critical/CB-2)", "Strict legal/governmental evidentiary bar for B2 requiring an explicit materiality finding of necessity or substantial enablement, plus confirmation delays", "Extreme rarity of peacetime mass-casualty (≥10 deaths) CBRN release events by non-state actors historically", "Anthropic's August 2026 Risk Report states: 'it is our expectation that we will declare near-future models to meet the RSP threshold for CB-2, or at least fail to rule it out', and says it expects to have the required mitigations ready at that time - a direct forward-looking statement from the resolving source.", "Current determinations are negative but close: Anthropic's Fable 5.1/Mythos 5.1 card (2026-09-01) says CB-1 met, CB-2 not met, with the gap located in novel ideation and strategic judgment; the Aug 2026 Risk Report called Fable 5 'our most capable widely deployed model nearing the CB-2 threshold'.", "OpenAI's GPT-6 Astra (2026-09-03) is Critical in cyber but only High in biological/chemical, with 0 of 4 Critical-bio evaluations above threshold and one regressing; OpenAI is separately reported to be rewriting the Preparedness Framework.", "Very high publication cadence gives many resolution opportunities: ~7 Anthropic system cards between Feb and Sep 2026 plus Risk Reports every ~6 months (next ~Feb 2027), and ~4-6 OpenAI system cards per year.", "Precedent that top-tier declarations are survivable: OpenAI declared Critical cyber, paused training for two weeks, and still deployed Astra.", "Accelerating capability trend: Amodei (2026-09-12) says AI has been 'advancing drastically faster' since summer 2026 via recursive self-improvement; Anthropic's roadmap puts the automated R&D threshold at 'plausibly, as soon as early 2027'.", "A political environment that now rewards candour and slowdown (Amodei's pacing campaign endorsed by Altman, Musk and Hassabis; Sanders pause bill; Thune/Cruz/Klobuchar duty-of-care legislation; calls for a superintelligence treaty), though offset by Anthropic's planned IPO and by a US administration publicly hostile to AI caution.", "Threshold drift risk: Anthropic revised CB-2 upward in May 2026 and OpenAI replaced its Critical-bio evaluations, so the goalposts have already moved twice in 2026.", "For B2, the binding constraint is not the underlying event but the government's willingness to make an explicit AI-materiality finding ('necessary to' or 'substantially enabled'); the question excludes the far more likely 'used'/'consulted' framing, and developer threat reports (Anthropic's five Sept 2026 bio-misuse cases, Google's Sept 2026 report) do not count.", "For B3, deliberate releases killing >=10 people run roughly 1-2 per decade globally (Tokyo sarin 1995, Ghouta 2013), none in the past year, and B3 additionally requires B2 to be satisfied for the same agent.", "OpenAI GPT-6 Astra (2026-09-03) is Critical in cybersecurity but only High in bio/chem; all GPT-5.5/5.6 models are High bio/chem — OpenAI's Critical bio bar is unmet", "Anthropic's August 2026 Risk Report (2026-08-14) says models meet CB-1 but explicitly do NOT yet meet CB-2 (novel CB weapons production), with 'substantial uncertainty'", "Rapid capability trajectory (METR time-horizon doubling; Anthropic says it may cross its AI-R&D threshold 'in the coming year') makes a top-bio declaration plausible within 5 years", "Both companies' frameworks have crossed top rungs steadily (Anthropic ASL-3 2025, CB-1 2026; OpenAI Cyber High then Critical 2026), supporting a high cumulative B1 hazard", "The 'pace the frontier'/pause push (Amodei essay 2026-09-12, backed by Altman/Musk/Hassabis) is a real risk of timeline stretching or disclosure delay", "No national-government/court confirmation of AI-materially-enabled Select-Agent or Schedule-1 acquisition has occurred; Anthropic's threat report is a developer document (excluded) for B2/B3", "B2's agent scope (Select Agents/Schedule-1/novel comparable) plus an explicit materiality finding is a stringent, rarely-issued standard, keeping B2 well below B1", "B3 (≥10 government-attributed deaths from a release of a B2-qualifying agent) is a compound, historically near-zero-base-rate tail" ], "would_update_on": [ "An explicit qualifying OpenAI or Anthropic threshold-met declaration.", "A national-government or court finding of covered-agent acquisition or synthesis substantially enabled by AI.", "Government-authored experimental evidence of real covered-agent synthesis with causal AI uplift and a near-term disclosure commitment.", "A linked release with at least ten deaths under national-government investigation or attribution.", "An implemented multi-lab slowdown lasting into 2028 or repeated strong negative results on relevant expert/end-to-end evaluations.", "The actual successor-definition rule or a policy revision clarifying whether changed thresholds preserve the original capability and harm requirements.", "Astra system card or next Anthropic system card stating 'cannot rule out' Critical bio / CB-2 (precursor to affirmation) — raise B1 near-term by 10+ points.", "UK AISI / US CAISI / SecureBio pre-deployment report concluding novel-agent uplift, undisputed for 14 days — raise B1 sharply.", "Anthropic announcing activation of ASL-4-level CB safeguards for a new model — raise B1.", "Next Anthropic system card again finding the same four CB-2 failure modes with little change, or a multi-quarter release freeze under a safety pact — lower B1 by 10+ points.", "An OpenAI/Anthropic threat report asserting a disrupted actor was part of an actual weapons program, or a developer confirming its model assisted a disrupted toxin/chemical plot — raise B2 by 10+ points.", "A casualty-causing chemical/biological attack with law-enforcement or developer statements about AI assistance — raise B3 substantially.", "Any system card or Preparedness/RSP evaluation rating a frontier model at the bio/chem Critical level (OpenAI) or CB-2 (Anthropic), or Anthropic shifting to 'we act as though our models meet CB-2' — would raise p_at_least[B1] sharply for subsequent horizons.", "Anthropic's ~Feb 2027 Risk Report or OpenAI's next flagship system card (expected ~early-mid 2027) classifying bio as at/near Critical or reporting saturation of bio safeguards.", "Any DOJ, FBI, HHS/CDC, or court filing that confirms an actor actually acquired/synthesized a Select Agent or CWC Schedule 1 chemical and states AI assistance was necessary or substantially enabling (would resolve B2 and raise B3).", "A further revision to the OpenAI Preparedness Framework or Anthropic RSP that narrows or renames the bio thresholds — strong evidence labs are managing around declaration, lowering p_at_least[B1] by 10+ points for later horizons.", "New VCT-style evaluations or autonomous-wet-lab demonstrations showing frontier models completing end-to-end synthesis cycles with lab tools without human intervention.", "An OpenAI or Anthropic system card or Risk Report that states Biological/Chemical Critical or Novel-CB/CB-2 is met.", "A further RSP/PF rewrite that raises, splits, or abandons the named novel/Critical rung without a 'met' statement.", "An undisputed SecureBio, UK AISI, US CAISI, or equivalent report that a named OpenAI/Anthropic model meets the expert-novel tier.", "A DOJ, FBI, HHS, CDC, or non-US agency/court document stating AI was necessary or substantially enabled acquisition or synthesis of a select agent, CWC Schedule 1 chemical, or comparable novel agent.", "Credible evidence of under-elicitation flipping Mythos/Astra expert-uplift from ~3/4 to the world-leading-expert line, or a major DNA-synthesis screening / cloud-lab failure tied to a listed agent.", "Anthropic's next Risk Report (expected ~Feb 2027) stating CB-2 is met, or language shifting from 'do not yet meet' to 'cannot rule out'/met for a Mythos-6-class model", "An OpenAI system card (GPT-6.5/7) rating biological/chemical Critical, or any of its three novel-design bio evaluations clearing the Critical bar", "Further revisions to Anthropic's CB-2 threshold or OpenAI's Preparedness Framework that raise/lower the bar or signal goalpost-moving", "A DOJ/FBI/foreign-government press release, indictment, or court filing describing an actual acquisition or synthesis of a select agent or Schedule 1 chemical with AI assistance characterized as necessary or substantially enabling", "Any release of a biological/chemical agent with casualties where officials attribute a role to AI assistance", "UK AISI / US CAISI / METR reports concluding frontier models provide expert-level novel-pathogen-design uplift (would pressure developers and shorten confirmation lag)", "OpenAI or Anthropic releasing an evaluation stating a model meets Critical (Bio/Chem) or CB-2 (novel bioweapons production)", "A DOJ or foreign court indictment/complaint alleging an actor synthesized a Select Agent or CWC Schedule 1 toxin with substantial AI assistance", "Significant breakthroughs in automated cloud laboratory robotics enabling fully autonomous end-to-end biological synthesis loops", "An Anthropic system card or the ~Feb 2027 Risk Report stating that a model meets (or 'provisionally meets' / 'we act as though it meets') the CB-2 threshold - would resolve B1 YES immediately.", "A further upward revision of Anthropic's CB-2 threshold (RSP v3.5+) or an OpenAI Preparedness Framework rewrite that renames, restructures, or removes the Critical biological-and-chemical category - would cut B1 by 10-20 points across all horizons.", "Evidence that Anthropic's stated CB-2 failure modes (open-ended novel ideation, strategic judgment, technical calibration in lab protocols) have closed in the next expert red-team round, or that OpenAI's four Critical-bio evaluations move materially toward threshold.", "A METR, UK AISI, or US CAISI report asserting that a named frontier model meets a top-tier bio threshold that the developer does not dispute within 14 days.", "Any DOJ/FBI/HHS/CDC or non-US equivalent statement, indictment, or court finding that an actor acquired or produced a Select Agent or Schedule 1 chemical and that AI assistance was necessary to or substantially enabled it - would resolve B2 YES.", "A US State Department Arms Control Compliance Report (or equivalent intelligence-community attribution) linking a state offensive BW program's production of a listed agent to AI/ML assistance.", "Announced pacing commitments that materially slow frontier release cadence (e.g. Anthropic delaying its next major generation into late 2027), or conversely an acceleration such as an Anthropic declaration that the automated AI R&D threshold has been crossed.", "Any deliberate release of a biological or chemical agent causing >=10 deaths, or any government attribution of a mass-casualty release, which would sharply raise B3.", "A developer stating a model MEETS (not 'cannot rule out') OpenAI Critical bio/chem or Anthropic CB-2", "A DOJ/FBI/HHS/CDC or non-US-equivalent statement or court document with an explicit AI-materiality finding on a Select-Agent/Schedule-1 acquisition", "A government-attributed release causing >=10 deaths from an AI-enabled CB agent", "A binding US/international pause or an RSP/PF revision that raises the top bio rung (would cut B1)", "Next Anthropic Risk Report (~Feb 2027) or the next OpenAI frontier system card, if either moves the CB-2/Critical-bio assessment" ], "forecasts": [ { "period_end": "2026-09-30", "p_at_least": { "B1": 0.0218, "B2": 0.0085, "B3": 0.0015 } }, { "period_end": "2026-12-31", "p_at_least": { "B1": 0.0734, "B2": 0.0224, "B3": 0.0028 } }, { "period_end": "2027-03-31", "p_at_least": { "B1": 0.1325, "B2": 0.0363, "B3": 0.0046 } }, { "period_end": "2027-06-30", "p_at_least": { "B1": 0.1946, "B2": 0.0519, "B3": 0.0067 } }, { "period_end": "2027-09-30", "p_at_least": { "B1": 0.2654, "B2": 0.0682, "B3": 0.0091 } }, { "period_end": "2027-12-31", "p_at_least": { "B1": 0.3296, "B2": 0.0862, "B3": 0.0116 } }, { "period_end": "2028-03-31", "p_at_least": { "B1": 0.3917, "B2": 0.1046, "B3": 0.015 } }, { "period_end": "2028-06-30", "p_at_least": { "B1": 0.4454, "B2": 0.124, "B3": 0.0183 } }, { "period_end": "2028-09-30", "p_at_least": { "B1": 0.4985, "B2": 0.1427, "B3": 0.022 } }, { "period_end": "2028-12-31", "p_at_least": { "B1": 0.5456, "B2": 0.1629, "B3": 0.0257 } }, { "period_end": "2029-03-31", "p_at_least": { "B1": 0.5849, "B2": 0.1815, "B3": 0.0298 } }, { "period_end": "2029-06-30", "p_at_least": { "B1": 0.6189, "B2": 0.1989, "B3": 0.0339 } }, { "period_end": "2029-09-30", "p_at_least": { "B1": 0.655, "B2": 0.2164, "B3": 0.0382 } }, { "period_end": "2029-12-31", "p_at_least": { "B1": 0.6805, "B2": 0.2348, "B3": 0.042 } }, { "period_end": "2030-03-31", "p_at_least": { "B1": 0.705, "B2": 0.252, "B3": 0.0464 } }, { "period_end": "2030-06-30", "p_at_least": { "B1": 0.7288, "B2": 0.2667, "B3": 0.0502 } }, { "period_end": "2030-09-30", "p_at_least": { "B1": 0.7512, "B2": 0.2832, "B3": 0.0545 } }, { "period_end": "2030-12-31", "p_at_least": { "B1": 0.767, "B2": 0.2975, "B3": 0.0584 } }, { "period_end": "2031-03-31", "p_at_least": { "B1": 0.7821, "B2": 0.3111, "B3": 0.0625 } }, { "period_end": "2031-06-30", "p_at_least": { "B1": 0.7938, "B2": 0.3242, "B3": 0.0664 } }, { "period_end": "2031-09-30", "p_at_least": { "B1": 0.8056, "B2": 0.3371, "B3": 0.0702 } } ] } ] }