Simon Willison — public-source brand diligence

Prepared: 2026-06-26
Creator: Simon Willison (@swillison on 10xN; usually @simonw elsewhere)
Recommended launch posture: Proceed with safeguards
Overall assessed brand risk: Low–Medium for a technical or developer campaign; campaign-specific risk can rise to Medium for AI-labor, privacy, anti-DEI, government, security, or exclusive AI-vendor work.

This is open-source reputation diligence, not a criminal, credit, employment, or consumer report. It intentionally excludes private contact details, family information, protected traits, medical information, and unverified gossip. “No issue found” means none was located in the sources reviewed; it does not prove an issue does not exist.

Executive assessment

Simon Willison has an unusually long, inspectable public record: a personal site dating to 2002, a documented role in creating Django, a founder exit, years at Eventbrite, a current open-source portfolio, and Python Software Foundation governance work. The core career claims checked here are supported by his contemporaneous writing and, where available, organizational sources. His public disclosures—on a standing page and in relevant posts—are materially stronger than the norm for technical creators: they identify sponsorship formats, vendor previews, free credits, paid appearances, hardware, foundation roles, and the boundaries of sponsor control.

No substantiated major personal misconduct, fraud, plagiarism, criminal allegation, regulatory action, or employment scandal was located in this review. That is a search result, not a universal clearance. The issues most likely to create launch friction are instead visible features of his current public work:

1. A dense conflict landscape. He covers AI vendors while receiving preview access, credits, event access, and, in isolated cases, payment or hardware. He also has GitHub Sponsors, banner/newsletter sponsors, consulting clients, a Fly.io-supported project, and a PSF board role. He discloses much of this, but the full current consulting-client and in-kind-benefit list is not public.
2. A strongly AI-forward public identity. In April 2026 he said roughly 95% of the code he produces is generated by AI. He also explicitly warns that quickly generated software may be unproven and that shipped code requires testing. This candor is a credibility strength, but it raises IP-provenance, quality, job-displacement, and audience-polarization questions for sponsored deliverables.
3. Editorial independence is a real constraint. His published policy says sponsors have no editorial influence and do not pay him to cover a specific subject. A campaign built around a scripted endorsement or undisclosed product placement would conflict with that policy.
4. Some public experiments are privacy-sensitive. In March 2026 he published a tool that uses up to 1,000 public Hacker News comments to have Claude profile a user. He described it himself as invasive, creepy, and mildly dystopian. He declined to quote another user’s generated profile in that post, which mitigates but does not erase the optics for privacy-sensitive brands.
5. Audience figures need normalization. The 10xN page displayed an unlabeled “57K” beside his profile on the review date, while his Substack said “over 63,000 subscribers” and YouTube showed about 7.81K subscribers. Those figures can coexist, but the channel, date, and methodology must be labeled in a media kit.

The most likely “bite us later” scenario is not a hidden historic scandal. It is a mismatch between a sponsor’s expectations and Simon’s already-public standards or positions: a sponsor expects message control; a competitor relationship was not surfaced; an AI-generated deliverable makes an unverified security/performance claim; a privacy brand discovers the Hacker News profiling experiment; or a campaign repeats a stale/unlabeled reach number.

Identity and official public footprint

Confirmed profiles and projects

• [Fact] 10xN’s public creator page listed “Simon Willison,” @swillison, and linked to his personal site on 2026-06-26.
• [Fact] His canonical biography and disclosure statement is simonwillison.net/about (accessed 2026-06-26).
• [Fact] Primary publishing site: simonwillison.net.
• [Fact] Code and open-source work: GitHub simonw. GitHub’s public user API showed 971 public repositories and 15,629 followers on 2026-06-26; both are volatile snapshot numbers.
• [Fact] Newsletter: Simon Willison’s Newsletter, which displayed “Over 63,000 subscribers” on 2026-06-26.
• [Fact] Video: YouTube @swillison, which displayed approximately 7.81K subscribers, 192,212 views, and 40 videos on 2026-06-26.
• [Fact] Current social accounts identified on his own about page include Bluesky, Mastodon, and legacy X/Twitter @simonw. His direct public Hacker News comments are under simonw.
• [Fact] Core product sites include Datasette, Datasette Cloud, and the llm CLI documentation (all accessed 2026-06-26).

Attribution cautions

• [Fact] 10xN uses @swillison, matching his YouTube handle, while several longer-standing accounts use @simonw. Use the canonical site to disambiguate.
• [Inference] Search results for “Simon” are noisy and frequently return unrelated entities. Negative-result searches should always use the full name plus a project or employer; a bare-name search is not reliable identity resolution.

How he got started

• [Fact — self-reported] In a 2012 account of his start in programming, Willison said he first programmed on a Commodore 64 as a child, later learned HTML and JavaScript, ran online gaming community/news/league sites, and learned Perl to automate that work.
• [Fact — self-reported] The same account says his first technology job was at Gameplay.com, initially editing a downloads section and later writing PHP/MySQL, before layoffs preceded his move into a computer-science degree at the University of Bath.
• [Fact — self-reported and contemporaneously documented] His 20-year blogging retrospective says he began the site on 2002-06-12 during his first year at Bath and did part-time work for Incutio. Early open-source work included IXR, an XML-RPC library later used by WordPress and Drupal.
• [Fact] In 2003 he undertook an industrial placement at the Lawrence Journal-World, where a Python web framework created for newsroom applications became Django. His 2005 launch post credits Adrian Holovaty and Willison with the initial development and Jacob Kaplan-Moss with joining the project; the Django project’s history FAQ independently places Django’s origin in the World Online newsroom in 2003 and its public release in 2005.
• [Context] “Co-creator of Django” is well supported. It should not be rewritten as “sole creator,” and the contribution should continue to be credited alongside the other early Django developers.

Public career and project timeline

Timeline qualifications

• [Fact/context] Eventbrite’s 2013 announcement said it would continue supporting Lanyrd. Willison’s 2022 retrospective says Lanyrd was eventually shut down because ongoing maintenance, security, and spam work could no longer be justified. That can disappoint former users, but no evidence reviewed shows fraud, concealment, or that Willison personally controlled Eventbrite’s later shutdown decision.
• [Fact/context] Datasette is mature enough to have years of production use, but the project’s 1.0 line was still publishing alpha releases in June 2026. Willison joked in a 2026-03-22 post that he had not been brave enough to ship 1.0 after many alphas. Avoid unqualified “GA,” “1.0,” or “enterprise-ready” wording unless the specific campaign has current substantiation.
• [Fact/context] Datasette Cloud was still described as a preview. Do not imply a generally available service, SLA, support tier, or customer scale that has not been documented.

Current work, businesses, and monetization

Products and projects

• [Fact] Datasette describes itself as an open-source tool for exploring and publishing data. On 2026-06-26 its site listed 44 tools and 154 plugins; those counts are volatile.
• [Fact] llm is a CLI and Python library that can use multiple proprietary and open model providers, save prompts/responses, generate embeddings, and call tools.
• [Fact] Datasette Cloud is a hosted commercial service in preview.
• [Fact] His tools repository describes itself as assorted useful tools “almost entirely generated using LLMs.” This is direct evidence of his extensive AI-assisted software practice, not evidence that every project is AI-generated.
• [Fact] His newsletter has a free weekly-ish edition; his about page says a paid monthly digest is available to GitHub Sponsors at the $10/month level or higher.

Disclosed revenue and material relationships

The following are public disclosures, not allegations:

• Banner/newsletter sponsorships. His disclosure page says he began weekly text-banner and newsletter sponsorships in February 2026 and that sponsors have no editorial influence. On 2026-06-26 the site carried a “Sponsored by Depot” banner using a 10xn.link redirect.
• GitHub Sponsors. Readers can support his work directly; higher tiers receive the paid digest.
• Fly.io. He says part of his Datasette Cloud work has been sponsored by Fly.io. In a 2026-01-09 post about Sprites, he disclosed that Fly sponsors some of his work while saying Fly neither requested that post nor provided preview access for it.
• GitHub. He is an unpaid GitHub Star and participated in the paid GitHub Accelerator in 2023. A 2023-11-10 post says GitHub invited him to speak as an Accelerator representative.
• Mozilla. His disclosure page says Mozilla’s MIECO program supported work in 2023–2024.
• AI-vendor access and benefits. He says OpenAI, Anthropic, Google/Gemini, and Mistral have invited him to previews under NDA/embargo, often including free API credits or event access.
• OpenAI paid preview appearance. In “Previewing GPT-5,” 2025-08-07, he disclosed that OpenAI paid for a couple hours of his time at an on-camera preview and that he signed an NDA and video release. He said OpenAI had no editorial control over his later coverage beyond the embargo.
• NVIDIA hardware. In a 2025-10-14 DGX Spark post, he disclosed receiving a preview unit valued at roughly $4,000, described it as his “first ever sponsored post if you like,” and said there was no cash payment or editorial control other than an embargo.
• Advertising/platform revenue. His disclosure page mentions occasional EthicalAds and X/Twitter creator revenue sharing.
• Consulting and training. He discloses ad hoc consulting/training clients and promises to disclose a relevant conflict, but does not publish a full current or historic client list.

Conflict assessment

• [Positive fact] Material relationships are frequently disclosed at the top of the relevant post, and the standing disclosure page is unusually specific.
• [Inference] Disclosure quality reduces deception risk but does not remove exclusivity, competitor, or perceived-bias risk. A user may see a positive model assessment without checking the standing page, and a foundation’s corporate relationship can create optics even when no money goes to the individual.
• [Fact/context] As a PSF director, Willison publicly supported the board’s decision to reject a proposed $1.5 million NSF grant containing an anti-DEI restriction (2025-10-27). He later welcomed Anthropic’s two-year $1.5 million sponsorship of the PSF; the PSF announcement describes an organizational, not personal, payment.
• [Inference] A sponsor may nonetheless perceive a conflict when he covers Anthropic while sitting on a foundation board funded by Anthropic. That relationship should be disclosed in an Anthropic-comparative campaign even though it is not personal compensation.

Audience and content profile

Audience evidence

Audience-claim conclusion: No contradiction is established. The figures appear to refer to different channels and/or dates. The risk is presentation ambiguity, not proven inflation.

Content mix and likely audience

• [Fact] The site mixes original essays, release notes, tutorials, conference notes, interviews, link-blog entries, and clearly formatted quotations from other sources.
• [Fact] Current recurring topics include generative AI and model releases, prompt injection and AI security, Python, Django, SQLite, Datasette, open source, data journalism, browser/web engineering, and AI-assisted development.
• [Inference] The core audience is technically sophisticated and likely to challenge vague benchmarks, unsupported superlatives, hidden affiliate relationships, and scripted praise. Claims should be reproducible and source-linked.
• [Inference] His value is credibility and depth with developers, not high-production lifestyle video. A technically substantive tutorial, benchmark, launch analysis, or transparent field test is better aligned than generic awareness copy.

Public positions likely to matter to sponsors

These are not misconduct. They are public stances that can make a brand pairing more or less suitable.

AI enthusiasm paired with explicit caution

• [Fact] In an April 2026 Lenny’s Podcast companion post, Willison said about 95% of the code he produces is now generated by AI and described very high daily output.
• [Fact/context] In the same post, he distinguished personal “vibe coding” from software shipped to users. He said creators should step back when bugs could harm other people and acknowledged that some quickly generated tools have documentation and tests but have not earned his confidence because he has not spent enough time with or used them.
• [Fact/context] He also discussed exhaustion, compulsive/gambling-like dynamics, uncertain labor impacts, and the need for responsible practice. This is more nuanced than an unconditional “AI replaces engineers” position.
• [Inference] AI-tool companies may find this alignment valuable. Developer-tools, labor, education, safety, and regulated-industry brands should expect scrutiny of the quality, provenance, and human-review process behind any deliverable.

Prompt injection and vendor security

• [Fact] Willison has long emphasized prompt-injection risk. His 2022-09-12 post introduced the “prompt injection” label while crediting Riley Goodside’s examples; later work includes his “lethal trifecta” formulation for dangerous combinations of private data, untrusted content, and external communication.
• [Context] Marketing should use “coined/popularized the term after public examples” rather than implying he discovered the entire vulnerability class alone.
• [Inference] He is unlikely to endorse unqualified “secure agent,” “prompt-injection-proof,” or “safe autonomous AI” claims without evidence.

Privacy and public-data profiling

• [Fact] On 2026-03-21, he published a tool that retrieves a user’s most recent 1,000 public Hacker News comments and asks Claude to profile that person.
• [Fact/context] He called the result invasive, creepy, startlingly effective, and mildly dystopian. He said he uses it to understand a commenter’s history and declined to quote another user’s generated profile in the post because doing so felt invasive.
• [Inference] A privacy, trust-and-safety, HR, recruiting, identity, or consumer-data sponsor could face criticism for pairing with this experiment unless the campaign directly acknowledges boundaries and consent concerns.

Platform, governance, and political-adjacent positions

• [Fact] In November 2022, after mass layoffs at Twitter including its accessibility team, he said Twitter’s priorities no longer aligned with his and moved his main social activity to Mastodon.
• [Fact] In 2025 he criticized xAI/Grok behavior while also recognizing model capability. The Associated Press quoted him discussing Grok’s tendency to consult Elon Musk’s views and its offensive “MechaHitler” episode (AP, 2025-07-12).
• [Fact] His 2025-10-27 PSF post strongly supported rejecting a US government grant with an anti-DEI clause.
• [Inference] These positions create higher fit risk for X/xAI, anti-DEI advocacy, or politically controlled government campaigns. They may be positive alignment for accessibility, open-source governance, and pro-inclusion brands.

Copyright and training-data concerns

• [Fact] He is broadly enthusiastic about LLMs but has repeatedly criticized closed model vendors for not disclosing training data; see “Closed model training,” 2023-06-04.
• [Inference] He may challenge broad “ethically trained,” “consented data,” or copyright-safe claims unless the sponsor can document them.

Debate style

• [Fact] His high-volume Hacker News participation sometimes uses blunt language. In a 2026-06-17 direct comment, for example, he called NFTs “very obviously stupid” while arguing LLMs are different.
• [Context] The reviewed examples were arguments about technology, not targeted harassment or protected groups. No pattern of threats or abusive conduct was identified.
• [Inference] This is a low-severity tone risk: a live social campaign should not assume every reply will use polished brand language.

Controversies, criticism, and adverse findings

1. No substantiated major personal scandal found

• [Search result] The reviewed source set did not surface a credible personal-misconduct scandal, lawsuit tied to misconduct, regulator action, fraud allegation, plagiarism finding, or public employer dispute involving Willison.
• [Limitation] This was not a paid court-record, sanctions, corporate-registry, or identity-verification search. Common-name search noise is substantial. Absence from this review is not proof of absence.

2. Lanyrd’s eventual shutdown

• [Fact] Eventbrite’s 2013 acquisition release said it would continue supporting Lanyrd.
• [Fact — Willison’s later account] His 2022 retrospective says the product was later shut down when maintenance, security updates, and spam defense could no longer be justified.
• [Inference/risk] Former-user disappointment and “acquired then closed” optics are plausible. No reviewed evidence supports treating this as personal wrongdoing. Severity: Low.

3. AI-generated software quality and provenance

• [Fact] Willison publicly says AI generates a large majority of his current code and labels one repository as almost entirely LLM-generated.
• [Context/rebuttal] He openly distinguishes prototypes from trusted production software, publishes tests/transcripts frequently, and urges stronger review for software that can harm users.
• [Inference/risk] Sponsored code, benchmarks, demos, and security claims could still contain model-originated defects, license/provenance uncertainty, or overstated maturity. This is not an allegation of infringement or defective code; it is a delivery-control issue. Severity: Medium for code deliverables; Low–Medium for editorial content.

4. Public software-security history

Datasette has published and patched security advisories. These are normal evidence of a maintained networked software product, not personal misconduct:

• [Fact] CVE-2021-32670 / GHSA-xw7c-jx9m-xh5g, published 2021-06-05: high-severity reflected XSS, fixed in 0.56.1 and 0.57.
• [Fact] CVE-2023-40570 / GHSA-7ch3-7pp7-7cpq, published 2023-08-22: low-severity database/table-name disclosure affecting 1.0 alphas, fixed in 1.0a4.
• [Fact] CVE-2025-64481 / GHSA-w832-gg5g-x44m, published 2025-11-05: low-severity open redirect, fixed in 0.65.2 and 1.0a21.
• [Context] Public advisories and fixes are a positive transparency signal. They also mean a sponsor must not use absolute claims such as “never vulnerable” or “unhackable.”
• [Unresolved allegation] An open third-party llm pull request #1424, filed 2026-04-28, uses “critical” language for several claimed issues. No maintainer validation or advisory was present in the reviewed record. Some described behavior may be intentional local-tool functionality. Treat the claims as unvalidated, do not repeat them as vulnerabilities, and ask for disposition before a security-centered campaign.

5. Privacy-sensitive profiling experiment

• [Fact] The Hacker News profiling tool is public and described by its author as invasive/creepy.
• [Context] It processes already-public comments, and he declined to quote another user’s generated profile in the post because doing so felt invasive.
• [Inference/risk] It can be recirculated without that nuance. Severity: Medium for privacy/HR/data-broker sponsors; Low otherwise.

6. Sponsor and vendor access can be framed as bias

• [Fact] Paid OpenAI time, an NVIDIA preview unit, vendor previews/credits/events, Fly sponsorship, GitHub Accelerator funding, and other relationships are disclosed.
• [Context/rebuttal] The standing policy rejects paid topic coverage and says sponsors lack editorial control. Relevant posts commonly contain prominent disclosures.
• [Inference/risk] Critics can still compile these benefits to question independence, especially during a model launch. Severity: Medium; likelihood depends on campaign.

7. Editorial-independence collision

• [Fact] His public policy says specific coverage cannot be purchased and sponsorship does not convey editorial influence.
• [Inference/risk] A contract requiring a positive conclusion, hidden talking points, prior approval of opinion, suppression of negative findings, or an undisclosed comparison would be inconsistent with the public policy and could produce a visible dispute. Severity: Medium and highly preventable.

Risk matrix

Credibility, IP, and disclosure review

Credibility positives

• A public archive spanning more than two decades permits unusually strong chronology checks.
• Core career claims are consistent with Django, Guardian, Y Combinator, Eventbrite, PSF, and project records.
• Posts commonly link source material, runnable code, issue threads, model transcripts, and corrections.
• Public security advisories include affected versions and fixed versions.
• His public disclosures name relationships that many creators leave implicit, including free credits, preview access, paid time, and hardware.

Credibility cautions

• High publishing velocity increases correction risk, especially during embargoed model launches. Every sponsor-supplied fact should have a source-of-truth document and timestamp.
• Model evaluations are snapshots. Providers silently change systems, rate limits, prompts, and pricing; preserve the exact model identifier, date, settings, and test inputs.
• “I built” can include extensive model-generated code. That is consistent with his disclosed workflow, but campaign contracts should specify what counts as authorship, review, and deliverable acceptance.
• Some content is a link blog or quote format, not independent reporting. Do not lift a linked claim and attribute its underlying reporting to him.

IP assessment

• [No adverse finding] No credible plagiarism finding or copyright judgment involving Willison was located.
• [Inference] Extensive LLM-assisted code creates unresolved industry-wide questions about training-data provenance and output similarity; the reviewed record does not establish infringement by him.
• Recommended control: For custom sponsored code or creative assets, require a repository/asset manifest, dependency licenses, source links, retained prompts where appropriate, human review, and confirmation that no sponsor confidential data was sent to an unapproved model provider.
• Recommended control: Do not require him to warrant facts outside his knowledge, vendor model provenance, or non-infringement of all generated output without a negotiated review process and realistic limitation of liability.

Disclosure assessment

Overall: Strong, with a campaign-level completeness gap.

The standing disclosure page is a major positive. The remaining issue is that readers do not necessarily visit it and it cannot list every active private consulting engagement. Each sponsored asset should therefore carry a compact, specific disclosure naming 10xN, the paying sponsor, compensation type, product access/credits/hardware retained, and whether the sponsor reviewed factual statements.

Claim ledger

Questions to resolve before contracting or launch

Ask these in writing and retain the dated answers:

1. What cash, equity, advisory, employment, consulting, affiliate, travel, event, hardware, API-credit, or preview-access relationships have existed in the last 24 months with the sponsor and its named competitors?
2. Is any current client confidential? If so, can counsel or an agency principal perform a conflict check without publicly naming that client?
3. Does he or an entity he controls retain the NVIDIA hardware or any other material review unit? Was any benefit taxable, returnable, or conditioned?
4. Which audience does the 10xN “57K” represent, what was its as-of date, and is it gross, active, or reachable? What are current newsletter open/click rates and trailing-90-day video/site results?
5. What factual-review workflow is compatible with his no-editorial-influence policy? Explicitly identify what the sponsor may correct and what it may not control.
6. Will the deliverable contain code, model output, benchmarks, screenshots, or third-party material? What human review, testing, license scan, and record retention will occur?
7. Will any sponsor confidential information or user data be sent to a model vendor? If so, which provider, plan, retention setting, region, and data-processing terms apply?
8. Does the campaign concern AI security or the llm CLI? If yes, what is the maintainer’s disposition of PR #1424 and are there unpublished advisories or embargoed reports?
9. Is the sponsor comfortable with his public positions on AI labor, X/xAI, DEI restrictions, model-training transparency, and public-comment profiling?
10. Does the campaign require category exclusivity? Define product category, named competitors, channels, geography, and duration narrowly enough to coexist with independent editorial coverage.
11. Who owns Datasette Cloud and the sponsored deliverables, and who is authorized to grant the promised rights? Do not infer this from GitHub ownership.
12. Are any performance, security, customer-count, uptime, “production-ready,” or “enterprise” claims planned? Identify the evidence owner and expiration date for each claim.

Recommended launch safeguards

Required

• Run a fresh conflict check no more than 48 hours before publication.
• Put a specific disclosure in every asset: payer, compensation/material benefit, relevant product access, and the boundary of sponsor review.
• Label every audience metric with channel, date, source, and definition; archive screenshots or platform exports.
• Preserve editorial independence in the contract. Permit factual, legal, confidentiality, and trademark review without guaranteeing praise or suppressing a truthful negative conclusion.
• Require primary-source support for product, benchmark, security, market-leadership, and competitor claims.
• For code/demos, require tests, human review, a dependency/license inventory, secret scanning, and sponsor acceptance in a non-production environment.
• Prohibit sending confidential sponsor/customer data to third-party models without written approval and suitable data terms.
• Use accurate career attribution: “Django co-creator,” “creator of Datasette,” and carefully scoped prompt-injection language.

Campaign-dependent

• For privacy/data/HR brands, review and explicitly account for the HN profiling experiment.
• For AI vendors, disclose relevant competitor previews, credits, payments, hardware, foundation funding optics, and ongoing consulting.
• For security campaigns, obtain an independent technical review and current vulnerability statement rather than relying on reputation.
• For X/xAI, government, or anti-DEI sponsors, treat the public-position mismatch as an up-front go/no-go question.
• For a Datasette Cloud campaign, use “preview” until the product itself changes that designation and separately substantiate support/SLA/customer claims.

Monitoring through launch

• Re-check the sponsor, creator, product, and campaign title for new posts/advisories on signing, final approval, and publication day.
• Monitor direct replies for factual errors, not merely negative sentiment. Correct quickly and preserve the original evidence.
• Keep a one-page response sheet for the five predictable issues: AI-generated code, vendor benefits, HN profiling, PSF/DEI stance, and audience metric definitions.

Unresolved evidence gaps

1. Private commercial roster: active consulting/training clients, equity interests, advisory roles, and confidential engagements cannot be reconstructed from public disclosures.
2. Audience quality: public follower/subscriber counts do not establish active reach, engagement, geography, demographics, overlap, or fraud-free traffic.
3. Legal/corporate verification: no comprehensive paid litigation, corporate-director, sanctions, insolvency, or identity-record search was performed.
4. Security posture: public advisories are not a current penetration test; unreported or embargoed issues may exist. The dramatic claims in llm PR #1424 remain unvalidated.
5. Deliverable provenance: public statements establish extensive AI use but not the provenance of any future campaign asset.
6. Ownership and insurance: public sources do not establish the contracting entity, E&O/cyber coverage, or who owns Datasette Cloud and every relevant trademark.
7. Historic archive completeness: a two-decade archive is large. This review sampled career milestones, disclosures, high-risk topic areas, public comments, product advisories, and external organizational sources; it did not manually read every post or comment.

Final recommendation

Proceed with safeguards; no present public-source blocker. Simon Willison’s public record, technical authority, and disclosure habits are strong. The launch should be designed around those strengths: evidence-heavy technical work, explicit disclosures, reproducible claims, and genuine editorial independence.

Do not proceed unchanged if the sponsor requires covert placement, guaranteed praise, broad category exclusivity, unqualified security/maturity claims, or control over his independent opinions. Pause and reassess if the pre-launch conflict questionnaire reveals a current competitor engagement, if audience substantiation materially diverges from the sales representation, or if a security-centered campaign cannot resolve the open llm allegation through competent technical review.