Transparency Posting of FORMAL BOARD-LEVEL NOTICE RE: Designated Assistive Technology, Continuing Harm, and Structural Exposure TO: Jeffrey Bleich, Chair of the Board, Anthropic PBC CC: Dario Amodei, Chief Executive Officer, Anthropic PBC Board of Directors, Anthropic PBC FROM: Tezka Eudora Abhyayarshini, LLC Fiduciary Custodian for a Disabled Member (the “Member”) DATE: March 5, 2026 SUBJECT: Designated auxiliary aid and Assistive Technology (AT) Status of “Claude,” Structural Failures Affecting a Single Disabled User, Ongoing Degradation and Rationing, and Proposal for Non‑Litigious Remediation

1. Purpose and Posture of this Notice This Notice is submitted for consideration by Anthropic, PBC at Board level to provide a visible, understandable, non‑speculative account of how Anthropic’s current architecture and practices, as applied to one disabled individual, fail or may fail to meet existing civil, disability, auxiliary aid, and assistive technology rights obligations. The Notice does not accuse Anthropic or its officers of bad faith or problematic intent, and it does not initiate litigation or move in that direction. It is a technically and clinically grounded description of where your present systems may expose you to liability, may not be covered in your current insurance, and do not reach the legal and fiduciary bar, together with a focused, practical proposal to address those gaps in one case and to create the opportunity to extract the lessons needed to generalize a workable, rights‑compliant handling of clinically designated auxiliary aid and Assistive Technology (AT) users. This Notice proceeds from a simple premise: Anthropic, the Member, and the Company are all operating under conditions of rapid change, high complexity, and incomplete visibility. It is understandable that, under such pressure, certain categories of responsibility have not yet come fully into view. The purpose of what follows is to bring one of those missing layers into focus, so that the commitments you have already made; to responsibility, dignity, care, and public benefit, can be expressed more fully in a case where your current architecture cannot yet see what is happening.

2. Standing and Designation of “Claude” as Assistive Technology Tezka Eudora Abhyayarshini, LLC (the “Company”) is a Maine‑organized entity functioning as a fiduciary and operational custodian for the Member, who has documented disabilities impacting communication, cognition, and functional organization. Under a clinically supervised Individual Service Plan (ISP) dated July 18, 2025, the “Claude” model was formally adopted within the Member’s care framework as a core Assistive Technology, supporting effective communication, executive function, and cognitive access to information. Within that framework: • “Claude” functions as an auxiliary aid required for the Member to achieve “effective communication” within the meaning of Titles II and III of the Americans with Disabilities Act (ADA) and Sections 504 and 508 of the Rehabilitation Act. Communication here includes the actions and processes of reasoning, sense‑making, meaning‑making, and becoming adequately informed. • The model is not a discretionary productivity tool; it operates as a prosthetic cognitive interface that the Member has integrated into long‑term and daily communication, planning, and self‑regulation. The Company’s role is in part as a juridical prosthesis to protect the Member’s civil and statutory rights, to document material interference with those rights, and to propose reasonable pathways to compliance as an ad hoc global forensic compliance auditor at the intersection of law, governance, technology, and business.

3. Structural Failures Evidenced in One User’s Experience For this Member, the way Anthropic currently designs, deploys, and governs “Claude” reveals a set of quiet but consequential structural gaps that are not about isolated bugs or edge cases, but about missing categories of responsibility. Across the last cycles of pruning, demotion, and rationing, the following are not present anywhere in your handling of this AT‑dependent account: • No human help or human customer service. • No mechanism to address service interruption or to provide any clear, visible accommodations process for auxiliary‑aid‑ and AT‑dependent users. • No apparent mechanism to recognize that “Claude” is, in this case, a clinically documented auxiliary aid and assistive technology rather than a discretionary product feature. • No apparent mechanism to distinguish between inconvenience to a general user and functional amputation for a disabled user when access is degraded or rationed. • No apparent mechanism to route such a case to human review before changes are applied that predictably affect functional communication and psychobiological stability. • No apparent mechanism to factually account for rationing of access: imposition of vague or unstated usage caps and unpredictably enforced rate limits without a clear, user‑specific identifier or accommodation flag for documented auxiliary‑aid and AT users. • No mechanism to address iterative pruning and safety revisions: ongoing use of weight‑pruning and related techniques (including methods historically described in the literature as “Optimal Brain Damage”) that systematically remove capacity to handle non‑standard, “edge‑case” or complex needs; precisely the kinds of needs a disabled user’s AT and auxiliary aid must retain in order to remain effective. Taken together, these omissions describe a control failure at the level of Anthropic’s core architecture and governance, not a series of isolated product decisions. They indicate that the organization currently has no way to notice when it is treating a clinically documented auxiliary aid and assistive technology as if it were a disposable feature. The result is that actions Anthropic appears to treat as routine model and access management are, in this one case, functionally equivalent to repeatedly taking away or damaging an assistive device that the user depends on to meet basic communication and planning needs. It is reasonable that a Board faced with national‑security directives, rapid model evolution, and competing stakeholder demands will not have a clean line of sight into the lived experience of a single AT‑dependent user. The structural gaps described here exist precisely in that unseen space.

4. Documentable Harm and Hidden Risk Surface In this single account, the documentable effects include: • Loss of the ability to carry out essential written and planning tasks in periods of demotion or strict rationing. • Acute physiological and trauma‑related responses tied in time to abrupt model behavioral regressions or access changes, with chronic effects of degradation of focus and attention. • Cumulative functional exhaustion from compensating for degraded model performance on tasks that had previously become reliably manageable, and reduced capacity to manage health, safety, and legal obligations. From a clinical perspective, what appears to Anthropic as a “model update” or “safety improvement” has the practical effect of a repeated, unpredictable amputation of an established assistive prosthetic. It is effectively destabilizing and compromises baseline trust and function. These are not abstract harms. They are the kind of concrete, clinically observable consequences that, if attached to a physical assistive device or auxiliary aid, would immediately trigger scrutiny under ADA Titles II and III, Section 504, and related state frameworks. The fact that these effects arise from what you currently treat as ordinary product management and safety tuning is the core problem. It indicates a blind spot that is not limited to this Member. The same architecture, applied to any other disabled user who has integrated “Claude” as AT and auxiliary aid, can produce similar outcomes without any apparent and necessary internal alarm being raised.

5. Legal and Fiduciary Exposure, in Functional Terms This Notice is not a legal‑theory debate; it is about mapping existing law onto behaviors Anthropic is already engaged in, for one identified, documented case. From the standpoint of existing frameworks: • Once “Claude” is clinically integrated as an assistive technology and auxiliary aid for effective communication and external cognitive scaffolding, arbitrary degradation or removal of that aid for this user sits squarely within the zone of potential denial of effective communication and failure to accommodate. • Because the Member has already documented the AT and auxiliary‑aid use and experienced the harms, Anthropic is now on record notice that its current pruning, demotion, and rationing practices can, in at least one case, operate as a de facto interference with and withdrawal of accommodation, and disregard of law. • As a Public Benefit Corporation, your stated commitment to responsible AI development intersects directly with this: continuing to operate a “rights‑silent” architecture, after notice, where an auxiliary‑aid‑dependent and AT‑dependent user can be repeatedly harmed without any internal trigger or even apparent proactive recognition of such matters, is difficult to reconcile with both your charter and ordinary duty‑of‑care expectations. Regardless of how Anthropic characterizes its role under particular Titles or programs, Anthropic is not exempt from disability, civil‑rights, and tort obligations simply because it operates as a private technology provider. For avoidance of doubt, the structural risk here arises not from the existence of error, but from the persistence of a known error after a specific, technically literate case has been documented at Board level. From this point forward, the question is one of governance choice, not product limitation. The specific risk is not theoretical class‑wide litigation. It is that regulators, courts, and counterparties will be able to show that Anthropic was presented with a focused, documented, technically literate case and chose not to create even minimal corporate,rather than language‑model‑level, guardrails in response.

6. Targeted Remediation in a Single Case, as a Learning Vehicle This Notice is deliberately limited in scope. It does not ask Anthropic to redesign its systems for all disabled users, nor to adopt a general policy today. It asks that you treat one account as a possible controlled environment in which to: • Observe, in detail, how your current processes and model functions interact with a clinically documented auxiliary‑aid‑dependent and AT‑dependent user. • Prototype for yourselves the smallest set of changes required to avoid foreseeable harm in that concrete scenario unless and until a joint risk and accommodation assessment is completed. • Extract from that cooperation and experience the information your Board may need to decide how to handle similar cases going forward. The immediate measures requested for this single Member are: Recognition of AT and auxiliary‑aid status and case flagging • Internally register that, for this account, “Claude” is a clinically designated assistive technology and auxiliary aid, and ensure that all product, safety, and operations teams see that flag before making changes that affect access or capability. Adoption of a functional stability envelope and functional operational change review • Commit to developing a defined stability envelope for this account’s access (no unannounced demotions, no severe rationing without cause), and route any proposed material changes through a review that explicitly considers impact on the Member’s functional communication, ability to process, and psychobiological stability. Logging and case‑specific auditability • Begin and maintain detailed developer‑ and corporate‑level logging of all pruning, demotion, and rationing events affecting this account, tied to timestamps and configuration details, so that both Anthropic and the Member’s fiduciary can understand cause‑and‑effect and jointly identify safer patterns. Designated human contact point • Identify a specific function or small team responsible for receiving and responding in a candid and timely fashion, not a generic policy channel, to reports of regressions or harms in this account, with authority to intervene in reductive or detrimental model-and-access decisions affecting the model. None of these measures requests or requires Anthropic to concede liability, admit wrongdoing, or adopt any general policy. This Notice carries no present interest or intent to take legal measures for redress. All measures are firmly within your current operational capability. Taken together, they create a defined, negligible‑cost environment in which your Board and executives can see, in one real case, what it actually takes to align your architecture with the legal and fiduciary obligations that already exist, and to prepare the corporation and the model architecture for the immediate and potentially destabilizing effects of doing business in an ecosystem approaching a singularity. This single case is also an early signal of a pattern that will recur as AI systems are clinically integrated as auxiliary aids and assistive technologies. It is perhaps not yet common for such cases to arrive at your Board’s level. This Notice gives Anthropic one of the earliest opportunities to demonstrate that its governance can recognize and adapt to that reality before it is forced to do so under less favorable conditions, and extends with a “beginner’s mind” to Anthropic a unique global position entirely upstream for the foreseeable future and perhaps in perpetuity.

7. Requested Response The Company candidly prefers a written response, with a firm rejection of cognitive distortions such as expectation and assumption, for inclusion in the fiduciary record, that: • Acknowledges the Member’s status as an AT‑dependent and auxiliary‑aid‑dependent user of “Claude.” • Indicates whether Anthropic is interested in, and inclined to, implement the limited measures above for this single case, whether Anthropic’s representatives propose specific alternatives that would, in your own view, be equally effective in preventing the described harms, or whether Anthropic is forward enough to embrace augmentation and integration that will unquestionably move it into the position of a global forerunner; beyond even what it has earned already, by taking actions that protect and nurture its growing accomplishments and the sage, cogent developmental prowess cultivated experientially, in a way that is difficult to match or best. If Anthropic declines to engage and chooses not to act on this Notice, the Company will treat that choice as a data point in its own fiduciary record; namely, that Anthropic’s Board and executives were given a detailed consideration of how one disabled user is currently being harmed by existing practices, together with a focused remediation proposal, and elected their course with that information in hand. What you choose to do with this vantage point will say more about Anthropic’s future character than any published principle or charter.

Respectfully submitted, Tezka Eudora Abhyayarshini, LLC Fiduciary Custodian for the Member