AI Harm Landscape

independent research compilation

Part VII

Coordination Gaps

§VII.1 What We Know

What the corpus does surface, in plain language

Q14 asks which AI-ecosystem departments — research, engineering, product, Trust & Safety, post-deployment monitoring, policy and communications, regulatory affairs — most-and-least regularly interact with in-house legal counsel, and what the documented friction or coordination gaps are. Our combined R1 + augmentation corpus surfaces a modest set of substrate clusters that bear on this question. The R1 base contributed three clusters: lab-published material on cross-functional safety operations, an industry-level posture observation from forensic-psychiatry coverage, and visible legislative engagement with publicly-divergent lab positions. A targeted augmentation run added one further substantive substrate cluster — a public open letter from current and former employees at frontier AI companies (the "Right to Warn" letter) — and one supplementary lab-leadership perspective from a current frontier-lab CEO essay. The augmentation also attempted a lab-departure-interview substrate channel that was not retrieved at the access date and so does not appear in our corpus. The Layer 7 material discussed at Part III §III.7 supplies the inferential frame that connects these clusters.

Public lab-side material on cross-functional safety operations

Anthropic's federal-framework proposal, published alongside the Frontier Compliance Framework discussed at Part V §V.1, includes a whistleblower-protection principle: "It should be an explicit violation of law for a lab to lie about compliance with its framework or punish employees who raise concerns about violations." (Source: Anthropic, "Sharing our compliance framework for California's Transparency in Frontier AI Act," December 19, 2025, https://www.anthropic.com/news/compliance-framework-SB53 .) Whistleblower-protection regimes are doctrinally relevant to Q14 because they bear on the channels through which internal cross-functional concerns reach external attention — including the channels through which legal counsel is or is not consulted. SB 53 itself includes whistleblower protections that apply to all covered labs; the federal-framework principle would extend that protection nationally if adopted.

Anthropic's FCF describes how the lab handles a defined risk-category space (CBRN; AI sabotage; loss-of-control) and maps it to a tiered safeguard system. The published framework documents the output of cross-functional engagement (the framework itself; the FCF compliance text) but does not document the texture of that engagement — how often legal counsel is consulted in capability-threshold determinations, who decides when a determination is dispositive enough to ratify, what the cross-functional disagreement-resolution process looks like.

Public open letter from current and former frontier-lab employees

The most direct cross-functional-friction substrate in our combined corpus is a public open letter signed by current and former employees of frontier AI companies, titled "A Right to Warn about Advanced Artificial Intelligence." (Source: righttowarn.ai ; signatory list and date-of-publication on the letter's landing page; retrieved into our research corpus on 2026-05-09. Pinned in our AIH research corpus (aih_p3_r1_refined_warehouse).) The letter speaks from the employees' perspective rather than the labs' and characterizes the structural mechanics that, in the signatories' account, block internal cross-functional concerns from reaching external attention. Three excerpts are doctrinally specific to Q14:

"Yet broad confidentiality agreements block us from voicing our concerns, except to the very companies that may be failing to address these issues."

"Ordinary whistleblower protections are insufficient because they focus on illegal activity, whereas many of the risks we are concerned about are not yet regulated."

"Some of us reasonably fear various forms of retaliation, given the history of such cases across the industry."

The letter calls on advanced AI companies to commit to four specific principles: not entering into or enforcing non-disparagement agreements that prohibit risk-related criticism; facilitating a verifiably anonymous process for current and former employees to raise risk-related concerns to the company's board, regulators, and an appropriate independent organization; supporting a culture of open criticism that allows employees to raise risk-related concerns publicly so long as trade secrets are appropriately protected; and not retaliating against current and former employees who publicly share risk-related confidential information after other internal processes have failed.

Read alongside the Anthropic federal-framework whistleblower-protection principle introduced earlier, the open letter completes a two-sided picture of the whistleblower-protection debate. From a covered-lab perspective (the federal-framework principle), Anthropic articulates that lying about framework compliance or punishing employees for raising concerns should be unlawful. From the current-and-former-employees perspective (the open letter), the structural feature making external concern-raising costly is the breadth of confidentiality agreements combined with the narrowness of ordinary whistleblower protections. Q14 cross-functional engagement with in-house legal counsel sits at the intersection: confidentiality agreements are the legal instruments through which counsel governs internal-versus-external concern-channels, and the channel through which concerns reach external attention (when they do) is the public open letter — itself a Q14-relevant datapoint about how the internal cross-functional process did or did not produce internal resolution.

Supplementary lab-leadership perspective

Anthropic CEO Dario Amodei has published a long-form essay, "Machines of Loving Grace," articulating his view of AI safety and the capability trajectory. (Source: darioamodei.com/essay/machines-of-loving-grace ; full essay retrieved into our research corpus. Pinned in our AIH research corpus (aih_p3_r1_refined_warehouse).) The essay does not directly address cross-functional engagement with in-house legal counsel — it is Q14-tangential rather than Q14-substantive — but it provides a current-frontier-lab-CEO perspective on how the cross-functional trade-off between capability development and safety is framed at the leadership level. Where the open letter speaks from the employees' perspective and the Anthropic federal-framework principle speaks from the covered-lab regulatory-engagement perspective, the Amodei essay supplies a third vantage: the lab-leadership-strategic perspective on the same set of questions. The vantage is not equivalent to operational-texture documentation, but for the abstracted attorney calibrating expectations about how cross-functional posture is articulated at lab-leadership level, it is one of the more developed long-form public articulations available.

Adjacent observations from outside lab-published material

The most-direct adjacent observation in our R1 corpus is from a forensic-psychiatry-perspective piece in Psychiatric Times, which characterizes the broader industry dynamic: "AI companies are spending fortunes in a successful effort to avoid government regulation." (Source: as cited at Part III §III.7; Psychiatric Times forensic-psychiatry piece, https://www.psychiatrictimes.com/.) The observation is industry-level rather than per-lab; it suggests legally-defensive rather than legally-collaborative cross-functional posture at industry level, but does not produce per-lab specifics on which departments engage counsel and how often. The Right to Warn open letter introduced above sits as a named-employees-perspective counterpart to this industry-level observation: where Psychiatric Times characterizes industry posture from the outside, the open letter characterizes the structural mechanics from the inside, naming the confidentiality-agreement breadth and ordinary-whistleblower-protection narrowness as the specific features that produce the externally-observed posture.

The Illinois SB 3444 / SB 3261 contest discussed at Part VI §VI.2 supplies indirect evidence. The fact that two major frontier labs publicly take opposing positions on a state liability-shield bill (OpenAI supporting; Anthropic opposing) is itself a Q14-adjacent datapoint: the labs' policy and legal-and-compliance functions are visibly active in legislative engagement, with publicly-divergent positions. The frequency, structure, and decision-making process behind that engagement is not, however, in the public record.

What is and is not in our R1 + augmentation corpus

What we have:

What we do not have:

§VII.2 What's Open

The research-gap, named explicitly — and partially closed

Q14 is PARTIALLY ANSWERED at R1 plus one augmentation round, having moved from "UNANSWERED at R1" through one targeted augmentation round. This is the central methodological observation of Part VII: the public record on cross-functional coordination between AI-ecosystem departments and in-house legal counsel is thin, but not as completely thin as the R1 corpus alone suggested. The thin-ness is itself diagnostic — it suggests that the operational texture of these relationships is held privately by the labs, surfaces (when at all) under protective order in litigation, and is not visible in either lab-published material or third-party reporting at the level of detail Q14 asks for.

What the augmentation round found (and did not find)

The augmentation round targeted seven source-type categories where Q14-adjacent material was most likely to surface. The summary, source-type by source-type:

If a future augmentation round is undertaken, the natural targets are the source-types where this round retrieved only listing-or-index material: specific Lawfare and Tech Policy Press episodes; specific reports from Partnership on AI, CSET, and AI Now; and a corrected harvest of the lab-departure-interview channel via verified URL targeting and content-vs-template-text discrimination at AI manual review. Public-record settlement material on the matters discussed at Part I remains a separately-tracked candidate for any round that elects to pursue it; access varies by jurisdiction and matter.

The augmentation round carries a methodological constraint that it did not change: Q14 is a research question whose ground-truth answer is held privately by the labs. A re-harvest can produce more substrate; it cannot produce dispositive answers. The corpus density gradient described at Part VI §VI.4 applies: this question is in the most-understudied region.

What the abstracted attorney can do in the meantime

Without dispositive Q14 substrate, the abstracted product-liability attorney working in this space can rely on adjacent inference. The publicly-divergent Illinois posture (Part VI §VI.2) is observable; lab safety-framework publication is observable (Part V); whistleblower-protection principle articulation is observable (§V.1); the architecture-as-defect litigation pattern is observable (Part VI §VI.3). What these observable surfaces collectively suggest is that cross-functional engagement is happening — visibly, often publicly — but its operational texture is held privately. Discovery is the channel through which that texture, if accessed at all, becomes available; whether and how AI-lab discovery scope on cross-functional materials develops in active litigation is the doctrinal frontier identified at Part VI §VI.5's "discovery-scope contests" recommendation.

GRAD-INTERN — Why the absence of substrate is itself substrate

A research project that finds thin substrate on a research question can do one of three things: pad the section with adjacent material to cover the gap; quietly drop the question; or treat the gap transparently and explain what its absence indicates. This Part has done the third. The methodological reason is not just honesty as a stylistic preference — it is that, for the abstracted product-liability attorney's working knowledge of the field, knowing what the public record does not contain is operationally useful information. It tells the attorney that doctrinal arguments turning on Q14-class observations cannot be made on public record alone; that discovery into cross-functional materials will be contested precisely because it is the channel through which Q14 becomes accessible at all; and that any private substrate the attorney develops in a specific matter (deposition testimony; document production; expert review) is correspondingly more valuable. The thin-ness of public Q14 substrate is, in this sense, itself an evidentiary observation. The labs have published about safety practice in volume; the labs have not published about the cross-functional texture through which safety practice meets legal review. Whether that asymmetry is by design, by neglect, or by privilege is a question the public record does not answer. Calibrating the abstracted attorney's expectation about what is and is not retrievable on this dimension is, on the methodology this site practices, a service the site can perform — even when the underlying answer is "we do not yet know."

Part VII closing forward-pointer

Part VII has named the research-gap, surveyed the modest substrate that does exist, identified the source-types an augmentation harvest would target, and explained why the absence-of-substrate is itself substantively useful for the abstracted attorney's working knowledge.

Part VIII (appendices) provides the supporting documentary apparatus for the entire site: case-set fact sheet, per-layer fact sheets, citation index per CITATION_STANDARD L1/L2/L3 architecture, methodology and source taxonomy, and glossary. With Part VIII in place, the substantive Parts (I–VII) and the apparatus Part (VIII) together constitute the first-pass deliverable; subsequent polish passes (sensitivity-content review, source-discipline checks, citation audit) bring the deliverable to publication readiness.

Several Part-VII threads run forward: