Automated Decision-Making and Artificial Intelligence Notice

§ 1. Plain-English Summary

In plain English: BeneVets, Inc. ("BeneVets") uses two automated systems that may affect you:

1. A crisis-keyword scanner that watches a few places where you type things into the Service and tries to notice if you might be in a mental-health crisis. If it does notice, BeneVets shows you the Veterans Crisis Line and other supportive resources. The scanner's guess about your mental-health status is treated as your most sensitive personal information. You can turn this scanner off in your Profile settings under Privacy. If you turn it off, the crisis-line button and the crisis banner are still available to you.
2. An AI text-summary tool that reads public government rules (like 38 C.F.R. and VA publications) and writes a plain-language summary that BeneVets staff then check before publishing on the website. At launch, this AI tool does not read anything you typed. It only reads public rules. Every summary links to the original government source.

This Notice tells you, in detail, what these systems do, what choices you have, and how BeneVets keeps them honest.

§ 2. Defined Terms

The following capitalized terms have the meanings given below. Terms not defined here have the meanings given in the BeneVets Privacy Policy.

"Automated Decision-Making Technology" or "ADMT" means any system, software, or process that processes Personal Information and uses computation to execute a decision, replace human decision-making, or substantially facilitate human decision-making, as that term is being developed by the California Privacy Protection Agency under Cal. Civ. Code § 1798.185(a)(16) and the regulations promulgated thereunder.

"Significant Decision" means a decision that results in the provision or denial of, or access to, financial or lending services, housing, insurance, education enrollment or opportunity, criminal justice, employment or independent contracting opportunities or compensation, healthcare services, or essential goods or services, consistent with the framework being developed under Cal. Civ. Code § 1798.185(a)(16) and parallel state regulations.

"Profiling" means any form of automated processing performed on Personal Information to evaluate, analyze, or predict aspects concerning an identified or identifiable natural person, including the analysis or prediction of a natural person's economic situation, health, personal preferences, interests, reliability, behavior, location, or movements, as that term is used in Cal. Civ. Code § 1798.140, Va. Code § 59.1-575, Colo. Rev. Stat. § 6-1-1303, Conn. Gen. Stat. § 42-515, Tex. Bus. & Com. Code § 541.001, and Or. Rev. Stat. § 646A.570.

"Large Language Model" or "LLM" means a type of artificial-intelligence model trained on text data that generates text output in response to text input.

"Sensitive Personal Information" or "SPI" has the meaning given at Cal. Civ. Code § 1798.140(ae), and includes, without limitation, Personal Information that reveals or concerns a consumer's health, including any inference drawn from other Personal Information about the consumer's health, mental or physical condition, or treatment, as set out at Cal. Civ. Code § 1798.140(ae)(1)(B) and (ae)(2).

"Consumer Health Data" has the meaning given at Wash. Rev. Code § 19.373.020, and means Personal Information that is linked or reasonably linkable to a consumer and that identifies the consumer's past, present, or future physical or mental health status, including mental-health-related information such as conditions, symptoms, diagnoses, treatments, and inferences thereof.

"System A" means the crisis-keyword scanner described in § 4.

"System B" means the LLM-summarized benefit-content tool described in § 5.

§ 3. Scope of This Notice

In plain English: This Notice covers the two automated systems above. It does not cover the rest of BeneVets's privacy practices. For the full picture, see the Privacy Policy.

This Notice is supplemental to, and incorporated by reference into, the BeneVets Privacy Policy. It addresses BeneVets's obligations under, among other authorities:

• The California Consumer Privacy Act, as amended by the California Privacy Rights Act, Cal. Civ. Code § 1798.100 et seq., in particular Cal. Civ. Code § 1798.121 (right to limit use of Sensitive Personal Information) and Cal. Civ. Code § 1798.185(a)(16) (Agency rulemaking authority for automated decision-making technology);
• The California Privacy Protection Agency regulations being developed at Cal. Code Regs. tit. 11, §§ 7150-7157 governing risk assessments and ADMT;
• The Washington My Health My Data Act, Wash. Rev. Code §§ 19.373.010-19.373.900;
• The Virginia Consumer Data Protection Act, Va. Code § 59.1-575 et seq., and the data-protection-assessment requirement at Va. Code § 59.1-580;
• The Colorado Privacy Act, Colo. Rev. Stat. § 6-1-1301 et seq., and the data-protection-assessment requirement at Colo. Rev. Stat. § 6-1-1309, together with the Colorado Artificial Intelligence Act, Colo. SB 24-205 (effective February 1, 2026), addressing high-risk artificial-intelligence systems and consequential decisions;
• The Connecticut Data Privacy Act, Conn. Gen. Stat. § 42-515 et seq., including the assessment requirement at Conn. Gen. Stat. § 42-522 (referenced herein as "CTDPA § 8" per the master context);
• The Texas Data Privacy and Security Act, Tex. Bus. & Com. Code § 541.001 et seq.;
• The Oregon Consumer Privacy Act, Or. Rev. Stat. § 646A.570 et seq.;
• The Federal Trade Commission Act, 15 U.S.C. § 45(a)(1), as applied to representations about artificial intelligence;
• The Substance Abuse and Mental Health Services Administration's Safe Messaging for Suicide Prevention guidance, as applied to automated handling of crisis-adjacent content; and
• Precautionarily, the European Union Artificial Intelligence Act, Regulation (EU) 2024/1689, in the event of any inadvertent EU exposure (noting that the Service is not targeted to EU residents).

§ 4. System A - Crisis-Keyword Scanner

§ 4.1 Plain-English Summary

In plain English: When you type into certain places on the Service (your symptom log, messages to a Helper, your profile narrative), a server-side scanner looks for words and patterns that might indicate you are in a mental-health crisis. If it sees something, the Service shows you the Veterans Crisis Line and other supportive resources. The scanner does not call 911. It does not contact your family. It does not tell anyone outside BeneVets. The actual words it noticed are stripped out before any BeneVets person sees the alert; only a label and a confidence score reach the moderation queue.

You may turn this scanner off at Profile > Privacy > "Limit use of sensitive personal information." If you turn it off, the scanner stops running on your content. The crisis-line button on every page and the crisis banner are still there for you.

§ 4.2 Technical Description

System A is a server-side keyword-and-pattern classifier. When a user submits text into a designated surface (see § 4.3), the classifier evaluates the submitted text against a curated list of crisis-indicative terms, multi-word patterns, and contextual rules, and returns:

• a label indicating whether the content is classified as potential-crisis, and
• a confidence score on a normalized scale.

Method strings - i.e., the specific words or phrases that triggered the classifier - are server-side redacted before any record reaches a staff-facing moderation queue or any log surface intended for human display. Only the label, confidence score, surface identifier, and timestamp are persisted.

System A operates entirely server-side. It does not perform inference on a third-party endpoint. It does not send user-entered text to any external provider.

§ 4.3 Categories of Content Scanned

System A scans content entered by the user into the following surfaces only:

1. The symptom-log free-text fields used in connection with C&P examination preparation (see Service Description § 2, item 6).
2. Helper-directed messages composed within the account-linking surface (see Service Description § 2, item 9), at the point of composition before transmission.
3. The free-text "profile narrative" or equivalent self-description field on the veteran profile (see Service Description § 2, item 3).

System A does not scan:

• Uploaded documents (including DD-214 envelopes in ProfileVault);
• VA form-field entries used for pro se form preparation;
• Email or push communications received from BeneVets;
• Any field not enumerated above.

§ 4.4 Output and Use of Output

The output of System A is used for the following limited purposes:

1. To surface in-product supportive resources to the user, including the Veterans Crisis Line (call 988 then press 1; text 838255; chat at https://www.veteranscrisisline.net/get-help-now/chat/; TTY dial 711 then 988), the VA Homeless Veterans Hotline, the Women Veterans Call Center, the VA Caregiver Support Line, and the SAMHSA National Helpline.
2. To enqueue a flag for review by a BeneVets crisis-trained moderation reviewer when that role is implemented (see § 4.9).
3. To inform internal bias-audit and quality-monitoring activities described in § 4.8.

The output of System A is not used:

• to contact law enforcement;
• to contact emergency medical services;
• to contact the user's family or Helpers;
• to disclose to any third party outside BeneVets;
• to deny the user any feature, benefit, or service of the Service;
• to make a Significant Decision about the user; or
• to train any artificial-intelligence model.

§ 4.5 Legal Classification

In plain English: California law treats a guess about your mental-health status as your most sensitive personal information. Washington law treats it as consumer health data. BeneVets agrees and treats it that way too.

The output of System A is an inference about the user's mental or physical health, condition, or treatment, and is therefore Sensitive Personal Information under Cal. Civ. Code § 1798.140(ae)(1)(B) and (ae)(2). The processing performed by System A constitutes automated decision-making within the meaning being developed under Cal. Civ. Code § 1798.185(a)(16) and Cal. Code Regs. tit. 11, §§ 7150-7157.

The processing involves Consumer Health Data within the meaning of Wash. Rev. Code § 19.373.020 and is conducted consistent with the consent and limitation requirements of Wash. Rev. Code §§ 19.373.030 and 19.373.040.

The processing constitutes Profiling under Va. Code § 59.1-575, Colo. Rev. Stat. § 6-1-1303, Conn. Gen. Stat. § 42-515, Tex. Bus. & Com. Code § 541.001, and Or. Rev. Stat. § 646A.570.

§ 4.6 Your Rights

§ 4.6.1 Right to Limit Use of Sensitive Personal Information (California)

In plain English: You can turn off the scanner. BeneVets will still let you use everything else on the Service. You will not be punished for turning it off.

Under Cal. Civ. Code § 1798.121, a California resident has the right to direct a business that collects Sensitive Personal Information to limit its use of that information to that which is necessary to perform the services or provide the goods reasonably expected by an average consumer who requests those goods or services. To exercise this right with respect to System A, navigate to Profile > Privacy > "Limit use of sensitive personal information" and set the toggle to the limited-use position.

Effect of exercising this right. When the limited-use toggle is set, System A does not execute on any text the user enters into any of the surfaces described in § 4.3. The user retains full access to the crisis-resource surfaces (the persistent crisis-line floating action button and the crisis banner) and may invoke them at any time on the user's own initiative.

§ 4.6.2 Right to Opt Out of Profiling (Other States)

A user who is a resident of Virginia, Colorado, Connecticut, Texas, or Oregon may exercise the right to opt out of Profiling in furtherance of decisions that produce legal or similarly significant effects concerning the consumer, under Va. Code § 59.1-577(A)(5), Colo. Rev. Stat. § 6-1-1306(1)(a)(III), Conn. Gen. Stat. § 42-518(a)(5), Tex. Bus. & Com. Code § 541.051(b)(5), and Or. Rev. Stat. § 646A.574(1)(e). BeneVets does not believe System A produces legal or similarly significant effects on users because System A is used solely to surface supportive resources and never to deny any feature, benefit, or service. The user may nevertheless exercise the right by setting the toggle described in § 4.6.1, which BeneVets honors across state regimes as a single equivalent control.

§ 4.6.3 Right to Meaningful Information About the Logic

A Colorado resident has the right under Colo. Code Regs. § 904-3, Rule 9.05, and a Connecticut resident under the equivalent CTDPA provision, to receive meaningful information about the logic involved in the Profiling. § 4.2 above (Technical Description) is provided to fulfill this right with respect to System A. Additional information is available upon request to the contact in § 12.

§ 4.6.4 Right to Non-Discrimination

Under Cal. Civ. Code § 1798.125, a business may not discriminate against a consumer because the consumer exercised any right under the CCPA. BeneVets will not deny the Service, charge a different price, or provide a different level or quality of the Service to any user who exercises the right described in § 4.6.1 or § 4.6.2.

§ 4.7 Risk Assessment

In plain English: State laws require BeneVets to write down a careful analysis of why this scanner is worth running, what the risks are, and how BeneVets reduces those risks. BeneVets is in the process of preparing that written analysis.

BeneVets is required to conduct, document, and maintain a data-protection assessment for the processing performed by System A under each of:

• Va. Code § 59.1-580;
• Colo. Rev. Stat. § 6-1-1309 and the implementing rules at Colo. Code Regs. § 904-3, Rule 8;
• Conn. Gen. Stat. § 42-522 (cited in the master context as "CTDPA § 8");
• Cal. Code Regs. tit. 11, §§ 7150-7157, as those regulations are finalized by the California Privacy Protection Agency; and
• Tex. Bus. & Com. Code § 541.105.

Status. A consolidated risk assessment covering System A is in preparation. Expected internal completion: [RISK ASSESSMENT TARGET COMPLETION DATE PENDING - set by Walter and counsel; see Question 1 in the cover memo]. The assessment will address, at a minimum: (a) necessity and the safety benefit of System A, including its alignment with SAMHSA Safe Messaging for Suicide Prevention guidance; (b) proportionality, including the limited surface scope at § 4.3, the user opt-out at § 4.6.1, and the 30-day retention cap at § 4.10; (c) mitigations, including server-side redaction of method strings, the prohibition on outbound disclosure at § 4.4, and the bias-audit corpus at § 4.8; (d) the rights of consumers; and (e) residual risks. Pending completion, BeneVets relies on the contemporaneous internal product-security review documented in the engineering repository. This Notice will be updated to reflect completion of the formal assessment.

§ 4.8 Bias and Fairness Considerations

In plain English: Keyword scanners can wrongly flag certain groups of people more than others. Veterans use military words that can confuse a scanner. BeneVets watches for that and tunes the scanner to be fair.

Keyword-and-pattern classifiers are susceptible to over-triggering on populations whose vernacular includes terms that look crisis-adjacent in general English but are routine within the population. Veterans, for whom military jargon, weapons-handling vocabulary, deployment narrative, and graveyard humor are common, are a known over-trigger population.

BeneVets monitors System A precision and recall against a maintained test corpus implemented at src/lib/moderation/bias-audit.ts. The bias-audit corpus is reviewed and refreshed on a semi-annual cadence. Terms and patterns are adjusted as the audit reveals drift. The output of the bias audit informs quarterly review by the moderation oversight committee described in § 9.

§ 4.9 Human Review and Crisis-Trained Reviewer

In plain English: The only automatic thing that happens when the scanner triggers is that the Service shows you supportive resources. If a person at BeneVets ever looks at the flag, that person will be trained for crisis content. That role is not staffed yet; until it is, no person looks at any flag.

The only automated action taken on the output of System A is the in-product surfacing of supportive resources described in § 4.4(1). No moderator action is taken on a System A alert except by a BeneVets crisis-trained moderation reviewer.

Status. The crisis-trained moderation reviewer role is not yet implemented at v1. Until that role is constituted, alerts accumulate solely in the bias-audit aggregate metrics described in § 4.8 and are not individually surfaced to any staff queue for case-by-case review. This Notice will be updated when the role is constituted to reflect the staffing, training requirements, escalation criteria, and review documentation standards then in effect.

§ 4.10 Retention

Trigger logs associated with System A - comprising label, confidence score, surface identifier, and timestamp, with method strings redacted - are retained for thirty (30) days from generation and then deleted. Aggregate, non-identifying metrics derived for bias-audit and quality-monitoring purposes may be retained longer in non-identifying form. Full detail appears in the BeneVets Data Retention Schedule.

§ 4.11 SAMHSA Safe Messaging Alignment

In plain English: When the scanner does its job, the Service follows safe-messaging rules - never sensationalize, never describe methods, always show help.

System A is designed and operated consistent with the Substance Abuse and Mental Health Services Administration's Safe Messaging for Suicide Prevention guidance. Specifically:

1. Method strings are server-redacted before any human-readable surface, in alignment with the guidance against describing or detailing methods.
2. In-product messaging surfaced on a System A trigger does not sensationalize, does not depict, and does not describe means.
3. Help resources - the Veterans Crisis Line at minimum, plus context-appropriate additional resources - are always promoted.
4. Content is presented in a manner intended to avoid harm to a vulnerable user.

§ 5. System B - LLM-Summarized Benefit Descriptions

§ 5.1 Plain-English Summary

In plain English: BeneVets uses an outside AI service to read government rules and write a plain-language summary. A BeneVets person checks the summary before it goes on the website. The AI does not read anything you typed. Every summary has a link to the real government source. AI can make mistakes, so do not rely only on the summary.

§ 5.2 Technical Description

System B is an automated content-generation tool. BeneVets transmits public regulatory and policy source material - including text from Title 38 of the Code of Federal Regulations, VA publications, and state benefit statutes - to a third-party Large Language Model provider. The provider returns plain-language summary text. BeneVets staff review the returned text for accuracy and fidelity to the source before any portion of it is published on a benefit-detail page within the Service.

§ 5.3 Inputs

The only inputs transmitted to the System B provider at v1 are:

• Public regulatory text (e.g., 38 C.F.R. provisions);
• Public VA publication text;
• Public state benefit statute text; and
• Staff-authored prompts and editorial instructions composed for this purpose.

No user Personal Information of any kind is transmitted to the System B provider at v1. No user account identifier, no user profile content, no user document content, no user free-text entry, no user IP address, no user device identifier, and no aggregated user telemetry is transmitted to the System B provider.

§ 5.4 Outputs and Editorial Process

The output of System B is candidate plain-language summary text. Before publication on a benefit-detail page, the candidate text is reviewed by BeneVets staff for:

• factual accuracy against the linked primary source;
• absence of any representation of legal advice or eligibility;
• conformance with the BeneVets editorial voice; and
• conformance with the plain-language prohibitions stated in the master engagement context.

Every System B-derived summary published in the Service displays a persistent link to the primary source from which it was summarized.

§ 5.5 Accuracy Disclaimer

In plain English: AI text can be wrong. Do not rely only on the summary. Always check the original source. Always talk to an accredited representative or a lawyer for advice about your specific situation.

LLM-generated summary text may contain errors, omissions, or misstatements. Users should not rely solely on the Service, including summary content generated with the assistance of System B, for benefit determinations. Users are encouraged to consult the primary regulatory source and an accredited Veterans Service Officer or other appropriately accredited representative under 38 C.F.R. § 14.629 for advice specific to the user's circumstances.

§ 5.6 Legal Classification

In plain English: Because the AI does not touch your personal information at launch, the heavier CPRA automated-decision rules do not apply to BeneVets's use of it for these summaries.

System B at v1 is an automated content-generation tool operated on public regulatory source material. Because System B does not process user Personal Information at v1, the consumer rights specific to automated decision-making under Cal. Civ. Code § 1798.185(a)(16) and Cal. Code Regs. tit. 11, §§ 7150-7157 do not attach to BeneVets's use of System B with respect to any individual user. The Federal Trade Commission Act, 15 U.S.C. § 45(a)(1), framework against deceptive representations of artificial-intelligence outputs nevertheless applies, and BeneVets's editorial and disclosure practices in this § 5 are designed to comply.

If BeneVets later uses an LLM to process user Personal Information - for example, to draft form pre-fills informed by the user's profile, to power a user-facing chatbot using user data, or to generate per-user content - BeneVets will:

1. update this Notice to disclose the new processing;
2. conduct a CPRA risk assessment under Cal. Code Regs. tit. 11, §§ 7150-7157, and parallel state assessments, prior to activation;
3. evaluate any high-risk thresholds under the Colorado Artificial Intelligence Act, Colo. SB 24-205, and precautionarily under the EU AI Act, Regulation (EU) 2024/1689; and
4. offer the user any opt-out or limitation right then required by law.

§ 5.7 Provider, Data Processing Agreement, and Data Residency

The System B provider is Anthropic, PBC (see Doc 04 Sub-processor List).

The Data Processing Agreement with the provider includes, at a minimum:

• a no-training-on-customer-data commitment, such that the content BeneVets transmits to the provider is not used by the provider to train, fine-tune, or improve any provider model; and
• a U.S.-data-residency commitment for the System B inference path used by BeneVets.

§ 5.8 Bias and Fairness Considerations

In plain English: AI can produce biased or wrong summaries of legal text. Staff review is BeneVets's main way of catching that.

LLM-generated summaries of regulatory content may reflect biases, inaccuracies, or framing inherited from the model's training. The primary mitigation at v1 is mandatory staff review of every candidate summary against the linked primary source before publication. Secondary mitigations include the persistent source link on every published summary (§ 5.4) and the accuracy disclaimer (§ 5.5).

§ 5.9 FTC Act § 5 Compliance

BeneVets does not represent System B-derived content as anything other than an editorial summary of the linked primary source. BeneVets does not represent System B-derived content as legal advice, as a determination of eligibility, or as a substitute for the primary source. These representational commitments are made in conformance with the Federal Trade Commission Act, 15 U.S.C. § 45(a)(1), and the FTC's enforcement posture on artificial-intelligence claims.

§ 6. Forward State - Future Uses of Artificial Intelligence

In plain English: If BeneVets ever adds AI features that touch your personal information, BeneVets will tell you here first and give you any opt-outs the law requires.

BeneVets may, in the future, expand the use of artificial intelligence within the Service. Potential expansions under consideration include AI-assisted drafting of form pre-fills informed by a user's profile, AI-assisted symptom-log structuring, and a user-facing assistant that draws on the user's profile to answer benefit-navigation questions. None of these are active at v1.

Before activating any such expansion, BeneVets will:

1. update this Notice to describe the new processing in plain English and in operative detail;
2. complete a CPRA risk assessment under Cal. Code Regs. tit. 11, §§ 7150-7157, and any parallel state risk or data-protection assessments required at that time, including under Va. Code § 59.1-580, Colo. Rev. Stat. § 6-1-1309, Conn. Gen. Stat. § 42-522, and Tex. Bus. & Com. Code § 541.105;
3. evaluate whether the processing meets the "high-risk artificial-intelligence system" or "consequential decision" thresholds under the Colorado Artificial Intelligence Act, Colo. SB 24-205, and apply that Act's developer-and-deployer obligations as applicable;
4. precautionarily evaluate the processing against the prohibited-practices, high-risk, and general-purpose AI obligations of the European Union Artificial Intelligence Act, Regulation (EU) 2024/1689, even though the Service is not targeted to the European Union;
5. confirm the contractual posture of any third-party LLM provider, including no-training-on-customer-data and U.S. data residency; and
6. provide users with any opt-out, limitation, appeal, or human-review right then required by law.

This Notice will not be a vehicle for retroactive expansion. New uses are not effective until disclosed here.

§ 7. Cross-References

This Notice is read together with, and incorporated into, the following documents:

• BeneVets Privacy Policy;
• BeneVets Children's Online Privacy Notice;
• BeneVets Helper / Account-Linking Terms;
• BeneVets Cookie Notice;
• BeneVets Sub-processor List;
• BeneVets Data Retention Schedule; and
• BeneVets Terms of Service.

In the event of a conflict between this Notice and the Privacy Policy on a matter addressed in this Notice, this Notice controls.

§ 8. How to Exercise Your Rights

In plain English: Here is how to use the choices described above.

§ 8.1 Turn Off System A (Crisis-Keyword Scanner)

Navigate to Profile > Privacy > "Limit use of sensitive personal information" and set the toggle to the limited-use position. The change takes effect on the next text submission. The crisis-line floating action button and crisis banner remain available regardless of the toggle state.

§ 8.2 Confirm Whether System B Has Been Expanded to User Data

System B does not process user Personal Information at v1. If this changes, that change will be disclosed in an updated version of this Notice. Users may consult the "Last reviewed" date at the top of this Notice and the Sub-processor List to confirm the current posture.

§ 8.3 General Privacy Rights

For the general suite of CCPA, VCDPA, CPA, CTDPA, TDPSA, OCPA, and Washington My Health My Data Act rights - including the right to know, the right to delete, the right to correct, the right to data portability, and the right to opt out of Sale, Sharing, or Targeted Advertising - see the "Your Rights" section of the BeneVets Privacy Policy and the verified-request submission process described there.

§ 8.4 Authorized Agent

A consumer may designate an authorized agent to exercise any right described in this Notice on the consumer's behalf, in accordance with Cal. Code Regs. tit. 11, § 7063, and parallel provisions in other state regimes. See the Privacy Policy for the agent-verification process.

§ 9. How BeneVets Keeps These Systems Accountable

In plain English: BeneVets watches the scanner for unfair patterns, tunes it, and reviews how it is working.

BeneVets maintains the following accountability practices for System A and System B:

1. Bias-audit corpus tests. The System A test corpus (src/lib/moderation/bias-audit.ts) is maintained and exercised on a semi-annual basis.
2. Periodic re-tuning. Terms and patterns in System A are adjusted in response to bias-audit findings, false-positive complaints, and clinical-guidance updates.
3. Editorial review. Every System B candidate summary is reviewed for accuracy against the linked primary source before publication.
4. Quarterly moderation oversight committee review (when constituted). A moderation oversight committee - to be constituted as part of the broader BeneVets governance program - will review aggregate System A metrics, bias-audit results, any user complaints, and any incidents on a quarterly cadence. Pending constitution of the committee, Walter (CEO and Executive Director) holds review responsibility and may consult Board members and counsel.
5. Notice updates. This Notice is updated when either system changes materially.

§ 10. SAMHSA Safe Messaging Alignment Statement

In plain English: BeneVets follows the recognized national guidance for how to handle suicide-prevention content safely.

BeneVets, Inc. operates System A consistent with the Substance Abuse and Mental Health Services Administration's Safe Messaging for Suicide Prevention guidance. Specifically:

1. Method strings are server-redacted before reaching any human-readable surface.
2. In-product messaging triggered by System A does not sensationalize and does not depict means.
3. Supportive resources - including the Veterans Crisis Line (call 988 then press 1; text 838255; chat at https://www.veteranscrisisline.net/get-help-now/chat/; TTY dial 711 then 988), the VA Homeless Veterans Hotline (1-877-424-3838), the Women Veterans Call Center (1-855-829-6636), the VA Caregiver Support Line (1-855-260-3274), and the SAMHSA National Helpline (1-800-662-HELP (4357)) - are promoted in context.
4. No System A output is published in any form that could harm a vulnerable user.

BeneVets does not operate the Veterans Crisis Line. The Veterans Crisis Line is a free, confidential service of the U.S. Department of Veterans Affairs.

§ 11. California AB 587 - Content Moderation Transparency Reporting

In plain English: A California law requires very large social media companies to publish a public report about how they moderate content. BeneVets is not a social media company and is not that large, so the law does not apply yet. If BeneVets ever grows into it, BeneVets will publish the report.

California Assembly Bill 587 (2022), codified at Cal. Bus. & Prof. Code §§ 22675-22681, imposes content-moderation transparency-reporting obligations on a "social media company" that generated gross revenue of more than one hundred million United States dollars in the preceding calendar year. As of July 4, 2026, BeneVets does not host user-generated content of the kind contemplated by the statute (no comments, no reviews, no forums, no user-to-user messaging surfaces other than the user-to-Helper composition surface) and does not meet the revenue or user-count thresholds for the statute to apply.

BeneVets has nevertheless prepared the internal framework necessary to publish a transparency report compliant with Cal. Bus. & Prof. Code § 22677 in the event the statute becomes applicable in any future period, and will publish such a report as required.

§ 12. Contact for ADM and AI Questions

Questions about this Notice, including requests for additional meaningful information about the logic of System A, should be directed to:

BeneVets, Inc. Attention: Privacy Team - Automated Decision-Making and AI 1441 Pomona Road, Suite 20, Corona, CA 92882 Riverside County, California Email: privacy@benevets.org

For California-specific rights requests, see the Privacy Policy "Your Rights" section.

§ 13. Changes to This Notice

This Notice is reviewed at least annually and on any material change to System A or System B. Material changes will be communicated in the manner specified by the BeneVets Privacy Policy. The "Last reviewed" date at the top of this Notice indicates the most recent review.

End of Document 15 - Automated Decision-Making and Artificial Intelligence Notice.