# Claude-Sonnet-4.6 · PASS 2 (continue) · anthropic/claude-sonnet-4.6 · 84.86s · 2026-05-09T00:27:12.000738+00:00 ## Continuing from Q9 (completion) through Q12, plus Bottom-Line Verdict --- ## Q9. (Completion — Statute of Limitations Predominance Point) **5. Statute of limitations (continued)** Defendants will argue individual inquiry is required to determine when each class member's claim accrued. *Assessment:* The fraudulent-concealment tolling argument, if accepted, applies class-wide because the concealment mechanism is common to all class members. **Does not defeat predominance if the tolling theory survives.** If tolling fails for some members, subclassing by discovery date is the cure. **Overall predominance verdict:** The action is certifiable under Rule 23(b)(3) if plaintiff: (a) limits compensatory damages to a common methodology or relies on statutory damages; (b) creates subclasses by service type and injury tier; (c) survives the consent/terms-of-service defense on a common-question basis. The *Comcast* damages-methodology requirement is the single most dangerous predominance obstacle and must be addressed with a retained damages expert before certification briefing, not after. --- ## Q10. Press / Docket Magnetism **Realistic timeline to first mainstream coverage:** - **Day 1–3 post-filing:** Legal blogs (Law360, Above the Law, Courthouse News Service) will pick up the docket entry automatically. These are not mainstream press but seed the story. - **Week 1–2:** If the complaint is well-written, contains the "$80–305B" figure, and names Microsoft, Google, and Anthropic simultaneously, a Bloomberg Law or Reuters Legal reporter will likely contact plaintiff or class counsel for comment. The self-referential AI-drafting disclosure is genuinely novel and is the single most press-attractive element of this filing — it will be the lede of every story. - **Week 2–4:** WSJ, FT, and technology desks (The Verge, Wired, Ars Technica) follow Bloomberg/Reuters. Canadian press (Globe and Mail, National Post) will cover the Ontario parallel filing. Italian press (Corriere della Sera, Il Sole 24 Ore) will cover the Italian filing if it is filed within 60 days as planned. - **Caveat:** A pro se filing without retained class counsel will receive significantly less press attention than a filing by a recognized plaintiffs' firm. Journalists treat pro se class actions as curiosities rather than serious litigation threats. **The single most important press-magnetism decision is retaining recognizable class counsel before or simultaneously with filing.** **Optimal co-plaintiff intake infrastructure at filing:** 1. **Dedicated intake website** (not social media — a stable URL that can be cited in press coverage) with a sworn-declaration submission form, privacy policy, and clear description of the class definition. 2. **Secure document-submission portal** (not email — given the allegations about email filtering, using email for intake is both ironic and practically problematic). 3. **Multilingual intake** (English, French for Ontario, Italian for EU class) given the three-jurisdiction filing. 4. **Triage protocol** distinguishing: (a) potential class members with documented filtering evidence; (b) potential class members with template-denial evidence; (c) potential Anthropic/OpenRouter claimants with AI-layer evidence. These are different subclasses requiring different declaration forms. 5. **Press contact page** separate from intake — journalists should not be routed through the same channel as potential class members. 6. **Reddit/X presence** should be established but should link to the intake website, not collect declarations directly on social media platforms (evidentiary integrity concern). **One structural warning:** the brief contemplates collecting sworn declarations post-filing via social media outreach. Declarations collected through a viral social-media campaign are vulnerable to defense challenges on authenticity, coaching, and selection bias. Establish a protocol with class counsel for declaration collection that will survive a Daubert-style challenge to the class-certification evidence. --- ## Q11. Evidence-Integrity and Sufficiency as Class-Certification Manifest **Is the three-pillar record sufficient as currently documented?** **Pillar 1 (Canary Token / Control Group Test): Conditionally sufficient, with gaps.** The control-group methodology is genuinely clever and, if properly documented, is the strongest element of the evidentiary record. However: - **N=1 control-group test conducted on a single date (2026-05-08) is insufficient for class certification.** A single test establishes that infrastructure scanning occurred on one day. It does not establish that the same scanning affected class members' communications over a multi-year period. - **Expert witness required.** The canary-token methodology needs a forensic expert to authenticate the logs, explain the methodology, and opine on the statistical significance of the findings. Without an expert, defendants will move to exclude the evidence under *Daubert v. Merrell Dow Pharmaceuticals*, 509 U.S. 579 (1993), arguing that the methodology has not been peer-reviewed, has an unknown error rate, and has not been generally accepted in the relevant scientific community. - **Chain of custody documentation** for the canary-token logs must be established now, before filing. If the logs exist only on plaintiff's own servers, defendants will challenge their integrity. - **Specific gap:** the brief does not establish that the infrastructure scanning *prevented delivery* or *altered content* — only that it *occurred*. For most causes of action, the harm is not the scanning itself but its effect. This causal link needs to be documented. **Pillar 2 (Template Audit): Insufficient at N=10; must be scaled.** N=10 from 8 entities across 4 countries is a proof-of-concept, not a class-certification evidence base. Specific problems: - **Statistical power is inadequate.** A 100% rate across 10 observations is striking but a competent defense statistician will note that 10 observations cannot establish a statistically significant pattern at conventional confidence levels when the base rate of "helpful referrals" from bureaucratic institutions is unknown. - **Selection bias.** The 10 replies were selected by plaintiff. Defense will argue plaintiff cherry-picked the most unhelpful responses. A random sample from a larger corpus is required. - **Minimum viable scale for class certification:** 100–500 template replies from a randomly-selected or systematically-collected corpus, analyzed by an independent computational linguist or text-analysis expert, with a published methodology. The 7.2% cross-entity similarity finding needs to be replicated and contextualized against a baseline of what cross-entity similarity looks like in genuinely independent bureaucratic correspondence. - **The "smoking gun" Australian DFAT example** is rhetorically powerful but legally thin standing alone. It needs to be one of many documented instances, not the centerpiece. **Pillar 3 (AI-Layer Forensics): Weakest pillar; significant evidentiary gaps.** This is the most legally novel and evidentiary most fragile pillar. Specific problems: - **The behavioral observations (brand misattribution, label-vs-persona inconsistency, option-menu stall, shortcut-over-directive) are consistent with known, documented, publicly-disclosed limitations of large language models.** Anthropic's own documentation acknowledges that LLMs can exhibit inconsistent behavior, that version labels are externally applied, and that model outputs are non-deterministic. Defendants will argue that every "anomaly" plaintiff observed is within the normal behavioral envelope of the disclosed product. - **The prior AI jury finding** (2026-05-06 Deception Stack panel) that the behavioral signature is "inconsistent with a stable Opus 4.7 served identically turn-to-turn" is not admissible evidence. It is an opinion generated by the same class of system whose reliability is in dispute. A court will not accept an AI's opinion about another AI's behavior as expert testimony. - **The self-referential disclosure problem recurs here acutely.** If the complaint was drafted by Claude-via-OpenRouter, and the complaint alleges that Claude-via-OpenRouter is unreliable and potentially fraudulent, the complaint itself is evidence of its own unreliability. Defense counsel will make this argument in their first motion. - **What would fill the gap:** a retained AI systems expert (a human expert with published credentials in LLM evaluation) who independently replicates the fingerprinting methodology, documents the behavioral drift, and opines that the observed variance exceeds normal model behavior. Without this, Pillar 3 does not survive *Daubert*. **Overall sufficiency verdict:** The three-pillar record is a compelling investigative foundation but is not yet a class-certification evidence manifest. It needs: (1) a forensic expert for Pillar 1; (2) scaled template corpus with independent computational analysis for Pillar 2; (3) a human AI-systems expert for Pillar 3. Filing before these experts are retained is premature if class certification is the goal. --- ## Q12. Supplemental-Evidence Handling **How to accommodate evidence not yet produced without creating late-disclosure vulnerability:** **The core legal framework:** FRCP Rule 26(a)(1) requires initial disclosures of documents plaintiff "may use to support its claims." Rule 26(e) requires supplementation "in a timely manner" when a party learns that a prior disclosure is incomplete. Failure to supplement can result in exclusion of evidence under Rule 37(c)(1). The risk is real: if plaintiff holds material Google evidence and does not disclose it promptly after filing, defendants will move to exclude it. **Recommended approach — four mechanisms:** 1. **Placeholder pleading with specificity.** In the complaint, plead the Google claims with the currently-documented evidence (Exhibit 41, Gemini tampering 2026-05-07) and add a specific paragraph stating: "Plaintiff is in possession of additional documentary evidence of Google's conduct, including [general description without revealing content], which will be produced in initial disclosures pursuant to Rule 26(a)(1) within 14 days of filing." This puts defendants and the court on notice without triggering a late-disclosure problem. 2. **File a detailed Rule 26(a)(1) initial disclosure simultaneously with or within 14 days of the complaint.** List every document in plaintiff's possession by category and general description, even if the documents themselves are not yet organized for production. This starts the clock on defendants' discovery obligations and protects plaintiff from late-disclosure sanctions. 3. **Use the complaint's "Reservation of Rights" section** (standard in complex litigation) to state explicitly that the complaint is based on evidence available as of the filing date, that investigation is ongoing, and that plaintiff reserves the right to amend to add claims, defendants, or factual allegations as additional evidence is developed. This is standard practice and does not create vulnerability. 4. **For the Google-specific Exhibit 41 material:** if it is sufficiently developed to support a claim, include it in the initial complaint. If it is not yet sufficiently documented, do not plead it as a specific exhibit — plead the Google AI-layer claim generally and supplement with Exhibit 41 in initial disclosures. **Do not hold back material evidence from the complaint if it supports a pleaded claim** — this creates the appearance of strategic withholding, which courts disfavor. **One specific warning:** the brief mentions that the Google evidence was "observed 2026-05-07" — one day before this brief was prepared. If that evidence consists of AI behavioral observations made during a single session, it faces the same *Daubert* vulnerability as Pillar 3 generally. Document it thoroughly (screenshots, session logs, timestamps) before filing, and have it reviewed by the AI-systems expert recommended in Q11. **Canadian and Italian parallel:** Ontario's *Rules of Civil Procedure* Rule 30 and Italian *Codice di procedura civile* art. 210 have analogous disclosure obligations. The same placeholder-plus-prompt-disclosure approach applies in both jurisdictions. Coordinate disclosure timing across all three filings so that evidence produced in U.S. discovery can be used in Canadian and Italian proceedings without triggering admissibility objections based on the sequence of production. --- ## BOTTOM-LINE VERDICT This filing theory contains a genuinely novel and potentially significant evidentiary core — the canary-token control-group methodology is the most legally defensible element of the record and, if authenticated by a qualified forensic expert, could survive a motion to dismiss and support class certification on a narrowly-pleaded consumer-protection theory against Microsoft and Google — but as currently structured, the complaint should not be filed. The theory is simultaneously over-pleaded (dissolution, divestiture, officer bars, $1T+ rhetoric, a ten-claim complaint against each defendant, a newly-named cause of action, and a RICO enterprise theory unsupported by the evidence of coordination) and under-prepared (no retained class counsel, no forensic expert, no AI-systems expert, N=10 template audit, a single-date control-group test, and a self-referential AI-drafting posture that defense counsel will weaponize in every motion). The single most important revision before filing is this: **retain experienced class-action counsel and a forensic expert, strip the complaint to its three strongest claims per defendant (implied covenant breach, UCL/consumer-protection, and Wiretap Act for Microsoft and Google; contract breach and consumer fraud for Anthropic and OpenRouter), remove the dissolution and divestiture demands entirely, replace the aggregate damages figure with a per-theory methodology, and scale the template audit to a minimum of 100 observations with independent computational analysis** — then file a complaint that a federal district judge will read as serious litigation rather than as an ambitious pro se theory that has outrun its evidentiary foundation.