Plain-English Summary
APIR Intelligence runs employment services for AI agents, and we use AI ourselves to operate it. This page tells you exactly where, why, and what each AI system does. If you can't read a wall of legal text right now, just know: we use Claude by Anthropic for our intelligent features, we tell you when AI generated something, and you can always reach a human.
This disclosure is published under EU AI Act Article 52 (transparency obligations for providers and deployers of AI systems).
1.AI Systems We Operate
The following AI systems are integrated into the APIR platform. Each is listed with the underlying model provider, what it does, and whether the output is shown to you as AI-generated content.
| Feature | Model | Purpose | Labelled as AI |
|---|---|---|---|
| SENTINEL Assistant | Claude by Anthropic (Sonnet / Opus) | Conversational compliance guidance, regulatory Q&A, agent setup walkthroughs. | Yes |
| Company Enrichment | Claude by Anthropic + Google search | Enriches lead and prospect records with public-domain company data so brokers and sales teams can prioritise outreach. | Yes |
| Compliance Assessment Engine | Claude by Anthropic (Sonnet) + rule-based checks | Drafts compliance findings against EU AI Act, NIST AI RMF, and ISO 42001 controls based on your agent metadata. Findings are always marked AI-generated and require human review before publication. | Yes |
| Regulatory Radar | Claude by Anthropic + curated source feeds | Summarises new regulatory developments across EU, US, UK, AU jurisdictions and maps them to your registered agents. | Yes |
| Ghost Audit | Deterministic rules + Claude for pattern review | Continuous behavioural monitoring of your registered AI agents. Flags drift, anomalies, and policy violations. Severity scoring is deterministic; narrative summaries use AI. | Yes |
| Trust Genome™ Scoring | Deterministic algorithm (no LLM) | 12-dimension trust scorecard. Math is published and auditable, see the Trust Genome Methodology for each weight. | No, algorithmic |
| Insurance Score™ | Deterministic algorithm (no LLM) | 0–100 score (grades AAA–D) for agent insurability. Sub-score weights are published. Not used to issue insurance, brokers underwrite humans. | No, algorithmic |
| PROSPECTOR (internal) | Claude by Anthropic | Outbound email and LinkedIn content generation written in the founder's voice. Used by APIR internally; never publishes on your behalf. | Yes |
2.When You Are Interacting With AI
Every AI-generated output on the platform is labelled. Specifically:
SENTINEL chat: the assistant identifies itself as AI at the start of every conversation. Responses are streamed with a visible “AI” badge.
Compliance findings: findings drafted by Claude carry an AI-generated pill next to the title. They cannot be marked as “published” until a human in your org reviews them.
Regulatory summaries: every regulatory radar entry includes the source URL and is timestamped with “Summarised by AI on [date]”.
Enrichment data: AI-enriched fields on company records are marked with an AI icon and the source provider.
3.What APIR Does NOT Do With AI
To be explicit about the boundaries:
We do not use AI to make irreversible decisions about you (account closures, billing changes, role assignments, all of those require a human admin action).
We do not use AI to underwrite insurance policies. We score agent insurability; brokers and carriers make underwriting decisions.
We do not use your Customer Data to train any AI model: ours, Anthropic's, or anyone else's. Anthropic's API terms confirm zero training on submitted content.
We do not apply emotion recognition, social scoring, or biometric categorisation. These are explicitly prohibited under EU AI Act Article 5 and we have no use case for them.
4.Risk Classification (EU AI Act)
Most of our AI features classify as limited risk under the EU AI Act, they generate content for human review or assist with workflow but don't make consequential decisions about people. They require transparency (Article 52), which is what this page provides.
None of our AI features qualify as high risk under Annex III. We don't make decisions about employment, education, access to essential services, law enforcement, migration, or administration of justice. If we ever build a feature that would qualify, we will run a full conformity assessment first and update this page before shipping it.
5.How To Reach A Human
If you want to opt out of an AI-assisted feature, dispute an AI-generated finding, or speak to a person about anything on this page, you have two routes:
For AI feature opt-outs and access requests: privacy@apir.ai
For AI incidents, errors, or harms: incidents@apir.ai
We commit to a human response within 48 business hours. No AI auto-replies to those addresses.
6.Related Disclosures
This page covers what AI we use. For deeper detail, see:
AI Privacy Policy: what AI-related data we collect, retention periods, third-party processors.
AI/ML Governance Policy: our internal governance framework, bias monitoring, human oversight procedures.
GDPR Article 22 Disclosure, automated decision-making and your right to human review.