AI Acceptable Use & Disclosure.

AI systems used

PursuitIQ's AI capabilities are powered by AURA (Adaptive Understanding and Reasoning Architecture), the platform's AI engine. AURA leverages the following foundation model infrastructure:

• Amazon Bedrock. The managed AI service layer through which all model inference is provisioned. Bedrock provides enterprise-grade security, data isolation, and contractual no-training guarantees.
• Anthropic Claude. The primary large language model used for natural language analysis, brief generation, chat interactions, and complex reasoning tasks. Accessed exclusively through Amazon Bedrock.
• Amazon Nova. Used for specific analytical tasks including document processing, structured data extraction, and high-throughput classification workloads.

All model access is provisioned through Amazon Bedrock within the AWS us-east-1 region. No Customer Data is transmitted directly to Anthropic or any model provider outside the Bedrock boundary.

What AI does

AURA powers the following capabilities within PursuitIQ:

• Pursuit scoring. Analyzing opportunity-to-company fit across multiple dimensions (past performance, capabilities, incumbency, budget, competitive landscape) to generate Pursuit Score.
• Brief generation. Producing structured opportunity briefs, executive summaries, and capture intelligence reports.
• Chat and Q&A. Answering natural language questions about opportunities, market conditions, competitors, and strategy.
• Competitive analysis. Evaluating competitive landscape, incumbent positions, and teaming possibilities.
• Budget and funding analysis. Interpreting appropriations data, obligation patterns, and Budget Runway projections.
• Bid/no-bid advisory. Synthesizing multiple data points into a recommendation framework for go/no-go decisions.
• Agent-driven workflows. PursuitIQ Agents automate research, monitoring, and analytical tasks within defined guardrails.

What AI does NOT do

AURA and PursuitIQ Agents are explicitly constrained from the following actions:

• AI does not make decisions. All outputs are recommendations and analysis. Bid/no-bid decisions, proposal submissions, pricing strategies, and resource allocations are made by humans.
• AI does not submit proposals. The Service never submits bids, proposals, quotes, or official responses to solicitations on your behalf.
• AI does not access external accounts. Agents cannot log into SAM.gov, email accounts, procurement portals, or any external system on your behalf.
• AI does not send communications. The Service never sends emails, messages, or notifications to external parties (government contacts, teaming partners, competitors) on your behalf.
• AI does not commit spend. No financial transactions, contract modifications, or resource commitments are initiated by AI without explicit human approval.
• AI does not operate autonomously. All Agent actions occur within defined permission boundaries set by your workspace administrator. Agents cannot expand their own permissions.

Data handling

When AURA processes your data for AI-powered features:

• AWS boundary. All AI inference occurs within the AWS environment (us-east-1). Customer Data processed by AI stays within the AWS boundary at all times.
• Per-org isolation. Each AI request is scoped to your organization. Your data is never co-mingled with another organization's data during inference, in context windows, or in any caching layer.
• Ephemeral processing. Model context is assembled per-request and discarded after response generation. No persistent context is maintained between separate sessions or shared across users from different organizations.
• No external transmission. Customer Data is never transmitted to model providers directly. All model access is through Amazon Bedrock, which provides a contractual boundary preventing data egress to model vendors.
• Logging. AI interactions are logged for audit, debugging, and usage metering purposes within your organization's scope. These logs are encrypted, access-controlled, and subject to the same retention policies as other Customer Data.

No-training commitment

This is an unqualified commitment:

This commitment is enforced at multiple levels:

• Contractual. Our agreement with Amazon Bedrock explicitly prohibits the use of customer request data for model training or improvement.
• Architectural. We use Bedrock's inference endpoints, which are architecturally separated from training infrastructure.
• Policy. Internal Company policy prohibits the use of any Customer Data for model development, evaluation datasets, or benchmark creation.

Your pursuit data, proposals, competitive intelligence, and strategic decisions remain exclusively yours. They do not improve the AI for anyone else.

Human oversight

AI outputs within PursuitIQ are designed as decision support, not decision replacement:

• Recommendations, not decisions. Pursuit Score, Bid Advisor, and Agent outputs are advisory. They surface relevant data and analysis to help qualified professionals make better decisions faster.
• Human-in-the-loop. No consequential action (bid submission, teaming commitment, resource allocation, proposal delivery) can be initiated by AI alone. Human approval is architecturally required for all actions with external impact.
• Override capability. Users can override, dismiss, or disagree with any AI recommendation without consequence. The system does not penalize human judgment.
• Transparency. AI-generated content is clearly labeled as such within the interface. Users always know when they are viewing AI-produced analysis versus source data or human-authored content.

Accuracy & verification

Specific accuracy considerations:

• Data currency. AI analysis is based on the data available at the time of the request. Government data sources may have delays between publication and ingestion.
• Scoring limitations. Pursuit Score reflects quantifiable factors and may not capture qualitative relationships, insider knowledge, or non-public agency preferences.
• Hallucination risk. Large language models may generate plausible but incorrect information. PursuitIQ mitigates this through grounding (Section 8) but cannot eliminate the risk entirely.
• Not professional advice. AI outputs do not constitute legal, financial, procurement, or professional advice. Consult qualified professionals for material decisions.

Grounding & guardrails

PursuitIQ employs multiple mechanisms to maximize AI output quality and minimize hallucination:

• Source citation. AI outputs cite the underlying data sources (SAM.gov records, USAspending data, FPDS entries, uploaded documents) that informed the analysis. Users can verify claims against source material.
• Deterministic guardrails. AI outputs are bounded by deterministic validation layers that verify factual claims against the structured data in our systems. Claims that cannot be grounded in source data are flagged or suppressed.
• Structured reasoning. AURA uses structured reasoning frameworks (not open-ended generation) for scoring and advisory outputs, reducing the surface area for hallucination.
• Confidence indicators. Where applicable, outputs include confidence levels or data-coverage indicators so users can calibrate their reliance on the analysis.
• Retrieval augmentation. AI responses are grounded in retrieved source documents and structured data rather than relying solely on model parametric knowledge.

Cost & metering

AI usage is metered and billed transparently:

• AURA credits. AI-powered features consume AURA credits, which are allocated based on your subscription tier. Credit usage is visible in real-time from your Workspace dashboard.
• Transparent pricing. Credit costs per operation type (chat query, brief generation, scoring refresh, Agent task) are published on the Pricing page.
• No hidden costs. You will never be charged beyond your allocated credits without explicit opt-in to overage billing.
• Usage dashboard. Real-time visibility into credit consumption by user, feature, and time period.
• Alerts. Configurable alerts when credit usage reaches defined thresholds (e.g., 80%, 95% of allocation).

Opt-out options

You maintain control over AI feature usage within your Workspace:

• Full AI opt-out. Workspace administrators can disable all AI-powered features, restricting the platform to deterministic-only analysis paths (structured scoring based on data rules, manual pipeline management, direct data access without AI synthesis).
• Feature-level controls. Individual AI features (chat, brief generation, Agents, AI-enhanced scoring) can be enabled or disabled independently.
• Per-user controls. Administrators can restrict AI feature access on a per-seat basis within the Workspace.
• Deterministic alternatives. Core platform functionality (opportunity search, pipeline tracking, data access, basic scoring) operates without AI and remains fully functional when AI features are disabled.
• No penalty. Disabling AI features does not affect your subscription tier, pricing, or access to non-AI functionality.

Changes to this policy

We will update the effective date above when this policy changes. Material changes — including changes to AI systems used, data handling practices, or training commitments — will be communicated to account administrators by email at least thirty (30) days before taking effect.