ABOS / sabirnetug-alzheimers — sabirnetug (ACU193) for early Alzheimer’s disease

Program analysis · bpiq_drug_id 14145 · prepared 2026-08 · USD · framework v5.5.0 · Coverage: CLEARED

Company context, financials, ownership, options and the full pipeline: ../company.md.

How to read this. Written for a reader who has not studied pharmacology or finance. Terms specific to this program are in the Glossary below; general terms are defined once in framework/04-glossary.md. Tags: [VERIFIED — source] / [UNVERIFIED] / [WEB ESTIMATE — source, date].

Glossary

Executive summary

• What it is (one sentence): A laboratory-made IgG2 antibody that binds small soluble clumps of amyloid-beta more than 650 times more tightly than it binds single molecules, and binds hardened amyloid plaque only weakly [VERIFIED — J Prev Alzheimers Dis 2025, DOI 10.1016/j.tjpad.2024.100005].
• The event and when (as disclosed): ALTITUDE-AD Phase 2 topline, guided “Late 2026” — a half-year, not a day. ClinicalTrials.gov records primary completion as 2026-10, which is consistent with a Late 2026 topline rather than in conflict with it. This document’s own readout window is 2026-11-15 to 2027-01-31, likeliest mid-December 2026 (see Readout) [VERIFIED — BPIQ fetch_company_drugs 2026-08-07; ClinicalTrials.gov NCT06335173, read 2026-08-07].
• The main reason it could work: The trial is the right size for the endpoint. Donanemab’s own Phase 2 hit the same primary measure at n=257; ALTITUDE-AD has 542 patients across three arms. Phase 1 showed dose-dependent proof the drug reaches its target in spinal fluid, movement in six separate disease markers, and zero cases of brain swelling or bleeding in the six patients carrying two copies of the highest-risk gene variant — the group that most often has to stop the approved drugs [VERIFIED — INTERCEPT-AD, DOIs 10.1016/j.tjpad.2024.100005 and 10.1016/j.tjpad.2025.100082].
• The main risk: The property that plausibly makes it safer is the same one that makes it clear less plaque, and the two drugs that have ever worked in this class are the two that cleared a lot of plaque. The oligomer hypothesis has never been validated in a well-powered human trial by anyone [VERIFIED — literature; the clinical consequence is UNVERIFIED and is exactly what this readout tests].
• What it means for the stock: A genuinely binary event on a company with roughly $103M of recomputed enterprise value that has already told its auditors there is substantial doubt it can continue as a going concern [VERIFIED — [../company.md](/analysis/ABOS) C.3 and C.4]. The modelled expected value sits about 44% above spot, and the modal outcome is still a loss of about three-quarters of the position.

0. Program-tier coverage — CLEARED

One follow-up call beyond the mandatory list was BLOCKED and is recorded here because it changes what can be asserted, not what the verdict is:

Company-tier coverage is in ../company.md C.0.

A. Scientific & clinical assessment

A.1 Mechanism of action and scientific rationale

What the drug is. Sabirnetug, also called ACU193, is a humanised monoclonal antibody — a laboratory-made protein engineered to stick to one specific target and given by drip into a vein. Its target is the soluble clumps of amyloid-beta that float in the fluid around brain cells, not the hardened plaques that show up on a scan [VERIFIED — J Prev Alzheimers Dis 2025, DOI 10.1016/j.tjpad.2024.100005].

How it works. Amyloid-beta is made normally in the brain. In Alzheimer’s disease it accumulates, and it does so in two forms: small soluble clumps called oligomers, and large hardened deposits called plaques. The scientific case for sabirnetug is that the oligomers, not the plaques, are what actually damages the brain — they stick to synapses, the junctions where nerve cells communicate, and destroy them. Sabirnetug was raised against those clumps specifically and binds them more than 650 times more tightly than it binds single unclumped amyloid molecules, of which the brain contains far more [VERIFIED — same source]. It also binds plaque relatively weakly, which is the whole argument and the whole risk at once.

How well the target is validated. Amyloid-beta as a target is validated at the level of the protein: two antibodies against it, lecanemab and donanemab, are approved and do slow decline [VERIFIED — Clarity AD (van Dyck et al., NEJM 2023) and TRAILBLAZER-ALZ 2 (Sims et al., JAMA 2023) headline results, taken as published and not re-derived from the primary papers in this sweep]. Open Targets confirms that both the amyloid precursor protein gene (ENSG00000142192) and Alzheimer’s disease (MONDO_0004975) exist as curated entities in its platform [VERIFIED — Open Targets search_entities, 2026-08-07]; the quantitative genetic association score could not be retrieved because the follow-up query was rate-limited across three attempts, so the strength of that link is [UNVERIFIED] here rather than asserted.

The oligomer hypothesis specifically has never been validated in a well-powered human trial by anyone. That is not a criticism of the science; it is the exact reason this readout exists. One piece of competitor-sponsored context is worth naming: a 2025 study from ProMIS Neurosciences (whose own PMN310 is an earlier-stage oligomer-selective antibody) reported that the antibodies which slowed decline in the clinic — lecanemab, aducanumab, donanemab — withstand monomer competition when binding brain-derived oligomers, while the failed pan-amyloid antibodies do not, and grouped ACU193 with the resistant set [VERIFIED that the paper says this — Alzheimers Dement (N Y) 2025, DOI 10.1002/trc2.70184; UNVERIFIED as independent support, because its senior authors are ProMIS employees and PMN310 competes with sabirnetug].

The exact scientific step this readout must prove. That neutralising soluble oligomers, without clearing most of the plaque, slows cognitive and functional decline by enough to reach statistical significance on the iADRS scale over 80 weeks. Every other question — safety, dose, formulation, commercial positioning — is downstream of that single result.

The honest scientific risk, stated plainly. The two anti-amyloid antibodies that have worked are the two that cleared a lot of plaque. The ones that cleared little — gantenerumab, solanezumab, crenezumab, bapineuzumab — all failed on clinical endpoints. In Phase 1, sabirnetug produced roughly 25% and 20% plaque reduction over three months at 60 mg/kg every four weeks and 25 mg/kg every two weeks respectively — in Centiloid terms, −18.2 and −18.3 Centiloids (p=0.01 in both cohorts) [VERIFIED — INTERCEPT-AD, DOI 10.1016/j.tjpad.2024.100005, full text read 2026-08-07]. Whether that rate continued over 80 weeks and reached the levels that the two winners achieved is not known, because no long-duration plaque data for this drug exists. It could extrapolate to substantial clearance or it could plateau. Both readings are available from the same fact, and anyone claiming certainty in either direction is guessing.

The shared-mechanism tension. An ex vivo study in mouse brain tissue found that lecanemab labels more cortical plaque and more cerebellar blood-vessel amyloid than sabirnetug, and encompasses a larger fraction of the total amyloid signal [VERIFIED — Alzheimer's & Dementia 2026, DOI 10.1002/alz.71509]. It is not independent: four of its six authors — Cline, Jerecic, Johnson and Siemers — are Acumen employees; only the first author and the senior author, Grenon and Lemere, are at Brigham and Women’s Hospital and Harvard [VERIFIED — PubMed author affiliations]. The finding still cuts both ways, which is why it is worth reporting: less vessel-wall binding is the mechanistic argument for less brain swelling, and less plaque binding is the mechanistic argument for less plaque clearance. The bull case and the bear case share one property.

The IgG2 point. Sabirnetug is an IgG2 antibody. Both approved competitors are IgG1. IgG1 antibodies recruit more of the immune system’s clean-up machinery, and that recruitment is the leading explanation for why brain swelling happens on this class. Acumen’s chief medical officer named this as the second thing the company will look at in the topline data after efficacy [VERIFIED — Acumen Q1 2026 earnings call, 2026-05-12]. It is a second, independent mechanistic reason to expect a better safety profile, and it does not depend on the plaque-binding argument. One cautionary Phase 1 detail belongs beside it: all five ARIA-E cases in INTERCEPT-AD occurred in women, and 21% of participants who received at least one 60 mg/kg dose experienced ARIA — small numbers, but not a flawless record [VERIFIED — INTERCEPT-AD full text, read 2026-08-07].

A.2 Clinical development plan, timeline, feasibility, resourcing

Known milestone dates for ALTITUDE-AD: first patient in 2024-05-08; last patient in 2025-03-26, giving a 10-month enrolment, completed ahead of the company’s own schedule; first participant dosed in the 12-month open-label extension 2025-11-17; primary completion 2026-10; topline guided Late 2026 [VERIFIED — BPIQ fetch_company_drugs note field 2026-08-07; ClinicalTrials.gov NCT06335173; company releases]. Database lock is not disclosed. The gap between a 2026-10 primary completion and a Late 2026 topline is a normal database-lock and analysis lag, not a disagreement between sources — the full timing judgement is in Readout, below.

Trials

Feasibility matrix

Resourcing sufficiency. Cash, burn, the term loan and the going-concern conclusion are in ../company.md C.3. The position for this program specifically is that the trial itself is fully paid for — enrolment closed in March 2025, the last patient’s 80-week visit falls in 2026, and quarterly research spending has already fallen from $25.3M to $16.5M year on year as the trial wound down. What the company cannot fund is what comes next. Its own stated runway ends in early 2027, a Phase 3 anti-amyloid programme costs several hundred million dollars, and the March 2026 placement was earmarked for the preclinical brain-delivery portfolio rather than for sabirnetug. This readout is therefore not a step towards a funded Phase 3; it is the event that determines on what terms the money to run one can be raised at all.

A.3 Target label, target product profile, strategic questions, designations

A.3a Target label. Early symptomatic Alzheimer’s disease — adults aged 50 to 90 with mild cognitive impairment or mild dementia due to Alzheimer’s disease, with amyloid pathology confirmed by brain scan or spinal fluid [VERIFIED — ClinicalTrials.gov NCT06335173].

A.3b Target product profile

A.3c Strategic Go/No-Go questions. The next decision for this asset is whether to run a Phase 3, so the pre-Phase-III set applies.

Pre-Phase-III (Go-to-Phase-III / registration):

A.3d Regulatory designations.

• FDA Fast Track, granted 2022-10-24 for Alzheimer’s disease. It grants more frequent written and in-person interaction with the FDA and eligibility for rolling review, meaning sections of a marketing application can be submitted as they are finished rather than all at once. It does not lower the evidence bar and it is not an approval pathway [VERIFIED — company announcement 2022-10-24, re-confirmed against the press feed 2026-08-07].
• No Breakthrough Therapy, Orphan Drug or Accelerated Approval designation is disclosed for sabirnetug [VERIFIED — no such designation appears in any 2024–2026 company release in the 207-item press feed].

A.4 Target product profile valuation matrix

Calibration examples, not asset claims: in oncology each incremental month of overall survival is worth roughly $150–300M in peak-sales potential; oral formulations command roughly 15–25% price premiums over injectables. Neither transfers directly to Alzheimer’s disease, where the binding constraint has been delivery infrastructure rather than price.

A.5 Evidence quality and endpoints

Endpoints

A.5b Key opinion leaders.

The CT.gov record for ALTITUDE-AD discloses no overall officials and no site investigators — every one of the 68 locations carries a null contact field — so the investigators below are named from the program’s own peer-reviewed publications instead, with the trial each name is verifiably attached to stated per row. Independent voices were found through PubMed and checked against both connectors for undisclosed relationships. Every attributed view carries a name, a date, a source and a conflict disclosure, or it is not recorded (rule 40); nothing here is collapsed into a sentiment score (rule 41).

Panel as of. 2026-08-07 — the date the investigator and independent-voice searches below were run.

Investigators

Being paid by the trial is a relationship with the sponsor by construction; the Conflicts column is for relationships beyond that one. A caveat that applies to every row: the INTERCEPT-AD paper’s own declaration of competing interests reads “None” despite four Acumen-employee authors, so this journal’s disclosure practice is thin, and an empty conflicts cell below records only that the named searches found nothing — never that nothing exists.

Independent voices

Named commentators on this program’s endpoint question who are not investigators on its trials and hold no disclosed relationship with the sponsor. In program.json every voice carries an explicit "conflicts": [] alongside the checks shown here.

Judgement

Dissent

Empty — the judgement above is MIXED, not SUPPORTIVE_CONTESTED, and the two skeptical voices are already recorded in full in the Independent voices table rather than repeated here.

B. Commercial assessment

B.0 Current treatment algorithm

A patient in the United States today who is losing their memory follows roughly this path.

1. Symptoms and first assessment. A primary-care doctor or a neurologist administers a short cognitive test such as the MMSE. This establishes that there is impairment; it does not establish the cause.
2. Confirming that it is Alzheimer’s. Before any anti-amyloid drug can be prescribed, amyloid has to be confirmed by a PET brain scan or a spinal tap. Increasingly a pTau217 blood test is used first to decide who is worth scanning. This step is a real access barrier: it costs money, it needs specialist equipment, and in Acumen’s own Phase 1 experience about 60% of scans ordered without a blood pre-screen came back negative [VERIFIED — Acumen Q1 2026 earnings call].
3. Symptom-relief drugs. Cholinesterase inhibitors such as donepezil and, in more advanced disease, memantine. These make symptoms a little better for a while. They do not change the disease. They remain the treatment most patients actually receive.
4. Disease-modifying anti-amyloid antibodies. For patients with confirmed amyloid and early disease only: lecanemab (Leqembi) by drip every two weeks, or donanemab (Kisunla) by drip every four weeks. Both slow decline modestly. Both require repeated MRI monitoring for brain swelling, both carry boxed warnings, and both need an infusion suite. This last requirement is the class’s binding constraint, and it is why realised sales are a fraction of what was forecast.
5. Later disease. Once dementia is moderate or severe, no anti-amyloid drug is indicated. Care is supportive.

Where sabirnetug would fit. Exactly at step 4, as a direct alternative to lecanemab and donanemab, in the same confirmed-amyloid early-disease population. It does not open a new line of therapy and it does not create a new patient pool. It would have to displace one of two incumbents that will by then have four to six years of real-world use, and its argument for doing so is a better balance of benefit and risk plus, eventually, an injection instead of a drip. There is one specific sub-population where displacement would be easiest: patients carrying two copies of ApoE4, who face the highest ARIA risk on the existing drugs and where sabirnetug’s Phase 1 recorded zero events in six patients.

B.1 Competitive landscape and positioning

Where this asset wins, and the single fact the thesis rests on.

It wins on one axis only: the possibility of comparable or better efficacy with materially less brain swelling, in a class where brain swelling is the reason most eligible patients never get treated. Everything else is a disadvantage. It is last to market, it is unfunded past early 2027, it has no commercial organisation, and its patents expire the year it would launch.

The single fact the thesis rests on is this: soluble oligomers, not plaques, are what damage the brain. If that is true, sabirnetug should produce class-leading efficacy at a fraction of the plaque clearance and a fraction of the ARIA. If it is false, sabirnetug is an antibody that clears less plaque than the drugs that work, and the Phase 2 will show it.

The competitor that matters most is not lecanemab. It is trontinemab (Roche). Trontinemab uses a transferrin-receptor shuttle to carry an anti-amyloid antibody across the blood–brain barrier, and its Phase 1b/2a produced 91–92% of patients turning amyloid-PET-negative at 28 weeks on the 3.6 mg/kg dose, with ARIA-E under 5%, one case mildly symptomatic — figures Roche reiterated at AAIC 2026 alongside evidence of clearance even in deep brain regions. It is in the Phase 3 TRONTIER 1 and 2 trials (~1,600 patients across 18 countries, started 2025), and new since the previous version of this analysis, Roche announced PrevenTRON at AAIC 2026 — a Phase 3 trial of trontinemab in cognitively unimpaired people at high risk, extending the programme into prevention [WEB ESTIMATE — NeurologyLive, Clinical Trials Arena and Roche releases, read 2026-08-07]. That combination — near-total plaque clearance and very low ARIA, now with a prevention franchise being built on top — is the outcome sabirnetug is arguing for by a completely different route, and Roche is three years ahead with a large balance sheet behind it. A sabirnetug result of, say, 25% slowing with 5% ARIA would be a scientific triumph and would still have to be sold against that.

Note also that Acumen’s own brain-delivery programme (bpiq_drug_id 20579 and 20578) uses the same transferrin-receptor idea, licensed from JCR. Management’s stated differentiation is that no competitor is shuttling an oligomer-selective cargo across the barrier [VERIFIED — Acumen Q1 2026 earnings call]. That is true and it is also several years and one funded IND away.

Calibration example, not an asset claim: oncology first-movers in novel mechanisms have shown roughly 3.2× higher peak-sales potential but about 1.8× higher development risk. Sabirnetug is a first-mover on mechanism and a laggard on time, which is the least commercially favourable of the four possible combinations.

B.2 Addressable market

Launch markets would be the United States first, then the European Union, the United Kingdom, Canada and Japan — the same markets where the approved drugs are sold and where infusion and MRI infrastructure exists.

B.3 Value and feasibility

B.3a Expected peak sales. Built bottom-up from the class’s realised revenue rather than from the epidemiology, because the class has spent four years missing epidemiology-based forecasts. All figures are annual net revenue at peak, conditional on approval, assume a 2030–2031 launch as a third or fourth entrant, and exclude the Enhanced Brain Delivery portfolio, any pre-symptomatic expansion, and any milestone or partnership income.

For context rather than as an input: third-party forecasts put Leqembi and Kisunla at roughly $2.9B and $2.3B of global sales by 2033 (GlobalData), and remternetug at ~$0.9B by 2033 [WEB ESTIMATE — GlobalData via Clinical Trials Arena, read 2026-08-07]. Those are forecasts for the market leaders. They are not passed through into the table above, because the same forecasting approach has been consistently high against realised sales, and rule 6 says a broken third-party figure is rejected rather than relayed.

B.3b Feasibility

B.4 Product-development risk

Framing questions.

1. Does the development plan support the target product profile claims? For the efficacy claim, yes: ALTITUDE-AD is the right design on the right endpoint in the right population. For the safety claim, only partly. The profile’s central promise is materially lower ARIA, and a 542-patient trial with roughly 360 patients on drug can show a large difference against the competitors’ published rates but cannot establish a precise rate, and cannot say anything reliable about the ApoE4-homozygous subgroup where the promise is most valuable — that subgroup will contain perhaps 30 to 60 patients across both active arms.
2. Will the identified risks affect the target product profile? Yes, one of them decisively. If plaque clearance at 80 weeks turns out to be modest and efficacy is weak, the profile collapses to “safer but doesn’t work”, which has no commercial value. The plaque-PET secondary is what will tell you which world you are in, and it will be in the same press release as the primary.
3. If a risk cannot be mitigated, is the asset still differentiated from competitors? Only if efficacy holds. A safety-only differentiation does not survive contact with trontinemab, which already shows ARIA-E under 5% alongside near-total plaque clearance — and is now extending into prevention with PrevenTRON.

Readout

Sources

The modelled estimate. Primary completion 2026-10 (registry), plus two to four months for database lock and topline analysis, gives 2026-12 to 2027-02. The company’s own guidance sits at the front of that range, which is consistent with a sponsor that finished enrolment early and has never moved a date: a database lock soon after the last Week-80 visit and a fast topline is the pattern its execution record supports.

Window

Basis. The last Week-80 assessments fall inside October 2026 (ctgov); the fastest credible database-lock-to-topline turnaround puts the earliest plausible announcement in mid-November 2026. Likeliest is mid-December: inside the guided “Late 2026”, consistent with primary completion plus roughly two months, and consistent with six unbroken reiterations of the same guidance. Latest is end-January 2027: a few weeks of ordinary analysis slippage past the guided year-end is retained despite the clean guidance record, but the modelled default’s February tail is discounted because this sponsor finished enrolment ahead of schedule and has held one date for a year. Precision is PERIOD because no source names a day or a month for the topline itself — “Late 2026” is a half-year, and the 2026-10 month belongs to the completion date, not the announcement. Confidence is MEDIUM: the sources agree and the guidance record is excellent, but the window rests on an undisclosed database-lock lag.

Disagreement. CONSISTENT — bpiq, company and ctgov all describe the same sequence (October completion, announcement in the following one to three months), and the modelled estimate overlaps the guided period. No source contradicts another; the modelled tail extending past “Late 2026” is lag uncertainty, not a conflicting disclosure.

Date slippage. Zero slips across six dated statements — the guidance has never moved.

Attribution

Status. CLEAN — computed by lib/clustering.mjs’s attributionFor over every has_catalyst row on this ticker’s pipeline, for bpiq_drug_id 14145, on 2026-08-07. ABOS carries exactly one has_catalyst: true row — this one — so the conflict set is empty by computation, not by assumption. The other three pipeline rows (17765, 20579, 20578) all carry has_catalyst: false and are excluded before the comparison begins, exactly as the computation defines. This block is stated even though it is empty because “nothing else lands in this window” is a finding the reader needs written down, not inferred from silence (rule 36).

Conflicts

Empty — exactly as attributionFor returned it.

Note. Blank — status is CLEAN.

Market and timing for this event

• Plain takeaway. A binary Phase 2 readout whose window opens in roughly three months, on a company with roughly $103M of recomputed enterprise value that has already declared substantial doubt about its ability to continue as a going concern. The largest holder owns 29.1% with a board seat and is under water on every tranche it has ever bought. No insider has ever bought a share in the open market. See ../company.md C.3, C.4 and C.5.
• Months to this catalyst. 3.3 months to readout.window.earliest (2026-11-15) from 2026-08-07, and 5.8 months to the window’s latest edge (2027-01-31). The disclosed date is still a half-year (“Late 2026”), so the window above is this document’s own judgement, at PERIOD precision — say so plainly: no source names a day. |
• Expected move around this event. The options chain cannot price it and cannot bracket it. ../company.md C.6 records three strikes in existence ($2.50, $5.00, $7.50) on a $2.28 stock, meaning there is no at-the-money option at any expiry; 153 contracts of open interest across the whole December 2026 expiry; and floor-pinned put volatilities. No figure is quoted from it. The move is built from the scenario anchors below instead. The one new positioning signal is 340 contracts of January 2027 call volume today — the first visible options interest spanning the readout window.
• Nearest comparable past reaction. There is none. Every one of the six measured reactions in ../company.md C.7 is a conference presentation, an enrolment milestone or a dosing announcement, and they span −12.60% to +19.38%. The closest analogue is 2025-03-19 (+19.38%), a real human topline — but of a formulation study in 28 healthy volunteers, not a controlled efficacy result. Acumen has never reported a controlled efficacy result in its life. The most useful thing C.7 says about this event is the newest row: on 2026-07-15 the stock fell 12.60% on cleanly positive preclinical news, which is the market saying it will price nothing but ALTITUDE-AD.
• Materiality. Dominant, per ../company.md C.2. It is the only pipeline row with has_catalyst: true, the only one that has ever produced human efficacy data, and the only one that can support a filing this decade. The stock-direction call below is consistent with that: both scenario ranges are multiples or fractions of spot, not adjustments to it. Attribution is CLEAN (above), so the move around this window belongs to this program alone.
• Date slippage. Zero. Six consecutive reiterations of “Late 2026” across twelve months, with no revision (Readout, above). It says nothing about whether the trial will work; it says the company has known when the data arrive for a year and has not had to move it.

Spot. $2.28, read on 2026-08-07, cited from ../company.md C.4. That is 45.2% of the way up a 52-week range of $1.19 to $3.60, and 86% below the $16.00 flotation price of July 2021.

Scenario prices

Expected value. Applying the 35% modelled probability below to the midpoint of each range: 0.35 × $7.75 + 0.65 × $0.875 = $3.28, which is +43.9% against the $2.28 spot. The sign does not flip anywhere inside the probability band: at 25% it is $2.59 (+13.7%) and at 48% it is $4.18 (+83.1%). This is arithmetic, not advice, and it is not a price target. It is also not a recommendation to own the position at any particular size: an expected value 44% above spot sits alongside a 65% probability of losing roughly three-quarters of the money, and those two facts have to be held together.

Run-up

Entry and exit

The exit rule resolves against the committed price cache. Today it resolves to nothing: the cache’s last bar is 2026-08-05 and the window opens 2026-11-15, so lib/runup.mjs’s resolveExit returns null for every rule — the correct answer, recorded as exit: null on the prediction, to be resolved once the cache reaches the window.

Predicted move and peak

Priority score drivers

Priority score. priority_score from lib/runup.mjs’s priorityScore, never hand-computed.

Priority score 11 · formula_version 1.0.0

Settlement. Left null at lock time. Settled only on an explicit user request, from a confirmed primary source, against the committed price cache. See 05-prediction-protocol.md § Run-up settlement.

Verdict

What I would do. Watch — and if you take a position at all, size it as a lottery ticket rather than as an investment.

Why. The expected value is meaningfully above spot and it stays above spot across the whole probability band, because the downside is bounded near $0.55 by modelled net cash plus a royalty-free Merck licence while the upside is a multiple. But the modal outcome is a miss, the miss costs about three-quarters of the position, and the company has already told its auditors there is substantial doubt it can continue as a going concern — which means even the good outcome arrives attached to an equity raise. The trial itself is well run: enrolment finished ahead of schedule, guidance has been reiterated six times over twelve months without a single slip, and the endpoint is the one donanemab won on at a smaller Phase 2 size. What is unproven is the science. Nobody has ever shown that neutralising soluble oligomers slows Alzheimer’s disease, and the property that makes this antibody plausibly safer is the same property that makes it clear less plaque. Since the last version of this analysis the external environment has hardened slightly against it: the April 2026 Cochrane review says the class’s effects are not clinically meaningful, and Roche extended trontinemab into a prevention Phase 3 at AAIC — neither changes the trial’s own odds, both compress what a merely-adequate result would be worth.

What would change this. One specific observable in each direction. Upward: 18-month amyloid-PET Centiloid data disclosed in or before the topline showing plaque reduction in the range the two approved drugs achieve. That would move sabirnetug out of the “modest clearance” bucket that every failed anti-amyloid antibody occupies and into the bucket both winners occupy, and it would justify a probability well above 35%. Downward: an equity raise announced before the readout. That would reset the miss-case floor on a larger share count and would signal that management does not believe it can raise on the data.

What to watch.

• 2026-08-12 — Q2 2026 results, now a scheduled date (announced 2026-08-05). Read three things: the cash balance against the modelled ~$91–95M, whether “into early 2027” changes, and whether the topline guidance narrows from “Late 2026” to a month or a venue.
• Any date — an 8-K, a shelf registration or an at-the-market facility. This is the single most informative filing that could appear before the readout, per the verdict above.
• 2026-10 — the ClinicalTrials.gov primary completion date for NCT06335173. If it moves, the clean twelve-month guidance record breaks and the readout window shifts with it.
• 2026-11-16 to 2026-11-19 — CTAD 2026, Boston. A plausible venue for the topline or a pre-topline disclosure — Acumen has presented at CTAD in each of the last three years — but no company statement names it, so it is a watch item and deliberately not a readout source.
• Throughout — Roche’s TRONTIER 1 and 2 and PrevenTRON news flow on trontinemab. A strong trontinemab result compresses sabirnetug’s commercial case even if ALTITUDE-AD succeeds.

Locked prediction

• Outcome-direction (will the readout succeed on its primary endpoint?): miss — probability 35%, band 25–48% [UNVERIFIED — modelled]. This is the probability that the settlement definition below is met: statistical significance on the iADRS primary in at least one active dose arm. It is below a coin flip for three reasons that compound. The oligomer hypothesis has never been validated in a well-powered human trial, so this is a first-in-class mechanism test rather than a repeat of something that has worked. Phase 1 plaque clearance of roughly 20–25% over three months is in the range that every failed anti-amyloid antibody occupied, and the Phase 2 doses were chosen for oligomer engagement rather than plaque clearance, with the top dose below the Phase 1 dose that produced those figures. And Alzheimer’s disease has a long record of well-designed, well-executed trials that failed. Set against that, three things keep the number from being lower: donanemab’s own Phase 2 hit this exact endpoint at n=257 against ALTITUDE-AD’s 542, so the trial is not underpowered by class standards; target engagement and six fluid biomarkers moved in the right direction in Phase 1 with p-values between 0.007 and 0.049; and execution has been faultless with zero date slips over twelve months. No third party has published a probability on this event. The nearest external markers are aggregator analyst consensus targets of $6.67–$7.34 against a $2.28 spot and Bank of America’s $8.00 Buy, all of which imply materially more optimism than 35%. They are named here rather than adopted. Unchanged from the superseded 2026-08-04 record: nothing that landed since — the Cochrane review, PrevenTRON — bears on the primary endpoint’s statistical odds.
• Stock-direction (which way do the shares move?): up — confidence low — window 2026-08-07 to 2027-02-28, basis: the readout window above (earliest 2026-11-15, latest 2027-01-31), plus four weeks for the reaction and any follow-on financing to settle. Called despite a below-even outcome probability, because the asymmetry is in the price rather than in the trial: the downside is bounded near $0.55 by modelled net cash plus a royalty-free Merck licence and two nominated brain-delivery candidates, the upside is a multiple, the expected value stays above spot at every point in the probability band, and the 29.1% holder with a board seat is anchored at $3.30 and cannot be a seller into weakness. Materiality is dominant per ../company.md C.2 — the only one of four pipeline rows with a disclosed catalyst — and attribution is CLEAN, so the move in this window belongs to this program alone. Confidence is low and not medium for three reasons: the company has declared substantial doubt about going concern, so a dilutive raise sits between the investor and the upside in both scenarios; ../company.md C.7 contains no precedent for how this equity trades on an efficacy result, because it has never had one; and a 65% probability of a roughly 75% loss is not a comfortable long at any size that matters.
• Scenario prices: positive $4.50–$11.00 · miss $0.55–$1.20
• Expected value: $3.28, +43.9% against spot $2.28 (arithmetic, not advice)
• Run-up: entry $2.28 on 2026-08-07, exit rule T-5 trading days before readout.window.earliest — predicted move +5–40%, predicted peak +20–75% around 2026-11 — priority score 11, formula_version 1.0.0. The score is capped by its own date-confidence gate: a PERIOD-precision window prices as nearly untradeable, whatever the other six drivers say. exit is null at lock — the price cache ends 2026-08-05 and the window opens 2026-11-15, so resolveExit correctly returns nothing yet.
• Settles on the first public disclosure of ALTITUDE-AD (NCT06335173) topline results. Positive definition: the topline reports a statistically significant slowing of decline for sabirnetug versus placebo on the pre-specified primary endpoint — change from baseline in the integrated Alzheimer’s Disease Rating Scale (iADRS) at Week 80 — in at least one of the two active dose arms (35 mg/kg or 50 mg/kg every four weeks), at conventional two-sided significance as pre-specified in the trial’s own statistical analysis plan. Anything else is a miss, including a numerical slowing that does not reach significance, and including a result that reaches significance only on a secondary endpoint such as CDR-SB. Source: Acumen’s own press release, or a peer-reviewed publication or conference presentation of the ALTITUDE-AD topline, cross-checked against the ClinicalTrials.gov record for NCT06335173.
• Locked: yes · Settled: no

Program data-quality flags

• The Open Targets follow-up association query remains BLOCKED across three attempts this sweep (immediate retry, then a retry after ~30 seconds), verbatim Rate limit exceeded for client: global — the third consecutive sweep of this program to record it. The mandatory search_entities call succeeded on the first attempt this time. The quantitative strength of the APP–Alzheimer’s genetic association is [UNVERIFIED] in this document rather than asserted at a number.
• ChEMBL returns no usable selectivity data for a biologic. CHEMBL5314772 carries no molecular properties, no structure and no bioactivity records, because sabirnetug is an antibody. The call confirms identity and development stage only. The >650-fold selectivity figure is the company’s own measurement reported in a peer-reviewed paper rather than an independent replication.
• The INTERCEPT-AD paper’s declaration of competing interests reads “None” despite four Acumen-employee authors and academic co-authors with documented competitor-trial relationships (A.5b). The journal’s disclosure practice is thin, and every empty conflicts cell in A.5b inherits that limit.
• PMC full text strips back matter. get_full_text_article on PMC13130801 (the Trials qualitative paper) returned the body without its Declarations section, so that paper’s own COI statement was not retrievable through this connector. Recorded in the A.5b conflicts-checked column rather than papered over.
• CT.gov discloses no investigators for NCT06335173. No overall official and no site contact is named on any of the 68 locations. The A.5b investigator table is built from the program’s peer-reviewed publications instead, and says per row which trial each name is verifiably attached to.
• The ex vivo lecanemab comparison is not independent. Four of the six authors of DOI 10.1002/alz.71509 are Acumen employees. Carried forward from the previous version, where this correction was first made.
• Errata exist on both key publications. DOI 10.1016/j.tjpad.2025.100213 (2025-05-30) corrects the main INTERCEPT-AD paper, and DOI 10.1016/j.tjpad.2025.100217 (2025-06-06) corrects the biomarker paper. Neither erratum’s content was retrievable in this sweep either, so what was corrected is [UNVERIFIED]. The figures quoted here are from the original papers, whose full text was read directly this sweep.
• INTERCEPT-AD site count disagrees between sources. ClinicalTrials.gov lists 17 locations; the publication says 15 study centres. Both are reported; neither is picked.
• The Phase 2 open-label extension is not separately registered on ClinicalTrials.gov under this sponsor. The sponsor’s registry footprint is exactly three trials (NCT04931459, NCT06511570, NCT06335173) and none is the extension, whose first patient was dosed 2025-11-17. Either it runs as an arm inside NCT06335173 without a separate record, or it is unregistered. Still unresolved.
• Competitor efficacy and ARIA figures are taken as published rather than re-derived. The 35% and 29% for donanemab, the 27% for lecanemab, and the 12.6% and 24% ARIA-E rates are the widely reported headline results of TRAILBLAZER-ALZ 2 and Clarity AD. The primary publications were not opened in this sweep, and that is the limit of the check.
• The JCR royalty rate could not be confirmed (the agreement covers the Enhanced Brain Delivery programme only, not sabirnetug). The Halozyme ENHANZE royalty — single-digit on net sales of the subcutaneous form — was confirmed this sweep from the 2023-11-05 agreement announcement, resolving a gap the previous version declared.
• Aggregator analyst-target figures disagree with each other (ChartMill: 12 analysts, $7.34; a second aggregator: $6.67 consensus; WallStreetZen: $8.00). The spread is quoted in the scenario anchors rather than one figure being picked.
• Company-level flags for ABOS are in ../company.md C.8.