Who this is for: Compliance officers at grocery retailers, food delivery platforms, retail technology vendors, pricing algorithm developers, and legal counsel advising food retail operations in Maryland.
Enforcement begins October 1, 2026. Maryland HB 895 prohibits the use of AI-driven surveillance pricing in food retail. Penalties: $10,000 per violation, up to $25,000 for repeat violations. Enforcement through the Maryland Attorney General and the Division of Consumer Protection. This is the first US law specifically targeting algorithmic price discrimination in the food sector.
Contents
1. Overview 2. Key Obligations 3. Obligation-to-Procedure Mapping 4. Detailed Procedure Cards 5. Quick Reference 6. Quick Start 7. References1. Overview
Maryland HB 895 is a first-of-its-kind US law that bans AI-driven surveillance pricing in food retail. "Surveillance pricing" is defined as the practice of using personal data, behavioral tracking, or algorithmic profiling to charge different prices to different consumers for the same food product based on individual characteristics rather than legitimate business factors.
The law responds to growing concerns about dynamic pricing algorithms in grocery stores and food delivery platforms that use consumer data -- purchase history, location, browsing behavior, loyalty program data, demographic profiling -- to set individualized prices. Unlike general dynamic pricing (which adjusts prices based on supply and demand), surveillance pricing targets individual consumers with personalized prices designed to extract maximum willingness to pay.
Key characteristics:
- Scope: Food retailers and food delivery platforms operating in Maryland, including online grocery services
- Prohibited practice: Using AI, algorithms, or automated systems to set individualized food prices based on personal consumer data or behavioral profiling
- Permitted practices: Coupons, loyalty program discounts uniformly available, volume discounts, clearance pricing, time-of-day promotions applied uniformly, and supply-demand adjustments not targeted at individuals
- Penalties: $10,000 per violation; $25,000 for repeat violations within 12 months
- Enforcement: Maryland Attorney General, Division of Consumer Protection
- No private right of action: Consumers cannot sue directly; enforcement is through the AG
2. Key Obligations
| Obligation | Requirement | Timeline |
|---|---|---|
| No surveillance pricing | Do not use AI to set individualized food prices based on personal consumer data or behavioral profiling | Continuous from October 1, 2026 |
| Pricing transparency | Prices displayed must be the same for all consumers for the same product at the same time and location | Continuous |
| Data use restrictions | Personal data collected through loyalty programs, apps, or tracking must not be used to increase prices for individual consumers | Continuous |
| Recordkeeping | Maintain records sufficient to demonstrate pricing practices comply with the law | Retain for AG investigation |
3. Obligation-to-Procedure Mapping
| HB 895 Obligation | SWT3 Procedure | What It Witnesses | Evidence Produced |
|---|---|---|---|
| No surveillance pricing | AI-FAIR.1 | Fairness attestation: pricing algorithm does not discriminate based on individual consumer data | Anchor with protected attribute (consumer identity), metric (price variance), threshold (zero individual targeting) |
| Pricing transparency | AI-TRANS.1 | Transparency disclosure: pricing methodology is uniform | Anchor with disclosure type (pricing policy), method (uniform pricing attestation), timestamp |
| Algorithm explainability | AI-EXPL.1 | Explanation of pricing factors used by the algorithm | Anchor with factors cited (supply/demand, clearance, volume), method, confidence |
| Data use restrictions | AI-DATA.1 | Data provenance: personal data is not used as pricing input | Anchor with dataset identifier (pricing inputs), provenance hash, data categories excluded |
| Bias audit | AI-FAIR.3 | Periodic audit: pricing outcomes do not vary by individual consumer profile | Anchor with groups tested, maximum price disparity, audit methodology |
| Recordkeeping | AI-AUDIT.1 | Audit log integrity for pricing decision records | Anchor with entry count, tamper detection result, retention period |
4. Detailed Procedure Cards
Pricing Fairness Attestation
HB 895 requires: Food retailers must not use AI or algorithms to charge different prices to different consumers for the same product based on personal data, behavioral profiling, or individual characteristics.
How SWT3 addresses it: The witnessFairness() call records the protected attribute tested (consumer identity/profile), the fairness metric (price variance across consumer segments), and the threshold applied (zero individual-targeted pricing). Regular fairness attestations create an ongoing record that the pricing algorithm treats all consumers equally for the same product.
AI-FAIR.1 anchors should appear at regular intervals (daily or per pricing cycle). Factor A identifies "consumer identity" as the protected attribute. Factor B shows the price variance metric (should be zero or near-zero for the same product/location/time). Factor C documents the fairness threshold. A non-zero variance for the same SKU at the same store and time indicates potential surveillance pricing.
Pricing Transparency Disclosure
HB 895 requires: Prices displayed must be uniform for all consumers for the same product at the same time and location.
How SWT3 addresses it: The witnessTransparency() call records the pricing policy disclosure, confirming that the retailer's pricing methodology does not incorporate individual consumer data. This creates a compliance attestation that can be presented during investigations.
AI-TRANS.1 anchors with disclosure type "pricing policy" prove the retailer has documented and attested to uniform pricing. Cross-reference with AI-FAIR.1 anchors to verify the attestation is backed by measurable fairness metrics, not just policy statements.
Data Input Provenance
HB 895 requires: Personal data from loyalty programs, apps, or behavioral tracking must not be used to increase prices for individual consumers.
How SWT3 addresses it: The witnessDataProvenance() call documents the data categories used as pricing inputs. By anchoring the dataset identifier and provenance hash, the system creates a verifiable record of what data feeds the pricing algorithm -- and what data is excluded.
AI-DATA.1 anchors for the pricing algorithm's input data. Factor A (dataset identifier) should reference aggregate market data, supply chain costs, or seasonal factors -- not individual consumer profiles. If personal data appears in the input manifest, it is a violation indicator. Cross-reference with the data pipeline architecture to verify no personal data pathway exists to the pricing engine.
Pricing Factor Explanation
HB 895 requires: Implicitly, retailers must be able to demonstrate that pricing decisions are based on legitimate business factors (supply, demand, cost, clearance) rather than individual consumer profiling.
How SWT3 addresses it: The witnessExplanation() call records the factors driving each pricing decision. Factor A identifies the explanation method, Factor B provides a confidence score, and Factor C lists the specific factors cited (e.g., wholesale cost change, seasonal adjustment, clearance timing). This creates an auditable record proving prices are set by business logic, not consumer profiling.
AI-EXPL.1 anchors for pricing decisions. Factor C should cite only permissible factors (supply/demand, wholesale cost, volume, clearance, seasonal). The presence of individual consumer attributes in Factor C (purchase history, location tracking, demographic data) would indicate a potential violation. Sample anchors across product categories to verify consistency.
5. Quick Reference
| Examiner Question | Where to Look |
|---|---|
| Does your pricing algorithm use individual consumer data? | AI-DATA.1 anchors for pricing inputs. Factor A should reference aggregate data only. No individual consumer identifiers or behavioral data should appear in the data manifest. |
| Are prices the same for all consumers? | AI-FAIR.1 anchors with price variance metric. For the same SKU/location/time, variance should be zero. Non-zero variance requires legitimate explanation (coupon, volume discount). |
| What factors determine your prices? | AI-EXPL.1 anchors. Factor C lists pricing drivers. Only permissible factors (supply, demand, cost, clearance) should appear. Individual consumer attributes are violations. |
| How do you audit pricing fairness? | AI-FAIR.3 anchors document periodic bias audits across consumer segments. AI-AUDIT.1 anchors verify log integrity for the pricing decision trail. |
| Do loyalty programs affect individual prices? | AI-DATA.1 anchors should show loyalty data is excluded from pricing inputs. Loyalty discounts are permitted only if uniformly available to all program members. |
6. Quick Start
pip install swt3-ai
# Initialize with the fairness monitoring profile
swt3 init --profile nist-ai-rmf --tenant YOUR_TENANT
# Run the demo to see fairness and transparency anchors
python -m swt3_ai.demo
# Or use TypeScript
npm install @tenova/swt3-ai
npx swt3-init --profile nist-ai-rmf
Full SDK documentation: sovereign.tenova.io/docs
Create a free account: sovereign.tenova.io/signup
7. References
- Maryland HB 895 -- Maryland General Assembly, signed 2026, effective October 1, 2026
- Maryland Division of Consumer Protection -- Enforcement authority
- FTC Surveillance Pricing Report -- Federal Trade Commission investigation into algorithmic pricing practices (2024)
- Colorado SB 26-189 Crosswalk -- comparable US state AI transparency law
- Idaho SB 1297 Crosswalk -- US state AI disclosure law
- SWT3 UCT Registry -- 103 AI procedures across 52 namespaces
- SWT3 Bidirectional Framework Crosswalks -- machine-readable JSON, 28 frameworks
- SDK Documentation -- Python, TypeScript, and 5 additional language SDKs