The contract was presented to city council as a public-safety investment: body-worn cameras (BWC, the recording devices officers wear on their uniforms) for the full patrol force, drones for tactical response, cloud storage for all video evidence, and an AI platform that would draft officer reports from footage and triage 911 calls by priority. One vendor, one contract, one signature. The price was approximately $45 million. The term was up to ten years. The procurement officer noted the vendor offered a 12 percent discount for signing a ten-year deal instead of five. The council approved it. Nobody on the dais asked what the agency would do in year eight if the vendor raised prices, changed the AI model in ways that affected report accuracy, or was acquired by a competitor with different data practices.

What the Bundle Buys and What It Costs

Bundled contracts for public-safety technology have become the dominant commercial structure in the law enforcement AI market. A single vendor offers cameras, drones, cloud evidence storage, AI report drafting, and AI dispatch analytics in one integrated package. The pitch is coherent and not dishonest: integration between systems is genuinely valuable. When the BWC footage uploads automatically to the cloud evidence locker, when the AI report-drafting tool can access that footage directly, when the CAD (computer-aided dispatch, the platform managing dispatch information) data feeds the same analytics dashboard as the officer reports, the workflow has real operational advantages over a patchwork of incompatible systems from six different vendors.

The cost of the bundle is not in the purchase price. The cost is in the dependency structure the bundle creates. When one vendor provides every component of the operational stack, switching any single component means renegotiating or terminating the entire contract. The agency cannot replace the AI report-drafting tool with a superior competitor's product without also replacing the cameras, the cloud storage, and the dispatch analytics. The vendor knows this, and the pricing structure reflects it. The upfront discount for the longer term is not a gift. It is a calculation that the vendor is willing to reduce the first-year margin because the subsequent years are locked.

The RMS (records management system, the platform storing and managing police reports and case files) and CAD integration compounds the dependency. When the AI platform is built to write directly into the RMS and read directly from the CAD, switching the AI platform requires either a new integration layer between the replacement product and the existing RMS and CAD, or replacement of the RMS and CAD as well. Each layer of integration is another chain in the lock-in structure. Each year the systems are used together, the operational habits, the report templates, the training materials, and the institutional knowledge become more vendor-specific.

A ten-year contract in a technology category that is changing as fast as AI in 2026 is not a long-term investment. It is a bet that the vendor you chose today will remain the best available option for a decade, and that bet has not aged well in public safety technology historically.

The Specific Risks of a Decade-Long Sole-Vendor Deal

A ten-year sole-vendor contract in public-safety AI creates five categories of risk that command staff should understand before any city council presentation on a bundled deal is made.

Policy Change Risk

The single most predictable risk in a decade-long public-safety technology contract is that the policy governing the technology will change before the contract ends. In 2026, agencies are navigating a policy environment that is shifting faster than any contract cycle. The King County, Washington, prosecutor's office barred AI-written police reports from use in prosecution. That policy change, had it occurred in year three of a ten-year AI report-drafting contract, would have left the agency paying for a tool it could not use for its primary purpose in that jurisdiction. The agency would still owe the vendor its remaining contract payments.

Policy changes that can affect a long-term public-safety AI contract include: prosecutorial policy changes regarding AI-drafted reports (King County is not the only jurisdiction where this has been raised), state legislation restricting AI use in law enforcement, city council resolutions following a high-profile AI-related incident, changes in the CJIS (Criminal Justice Information Services) Security Policy that affect how vendors must handle agency data, and changes in how courts treat AI-assisted evidence following appellate rulings. A ten-year contract has no mechanism to respond to any of these changes unless the contract specifically includes a policy-change termination clause. Most standard vendor contracts do not include one.

Technology Change Risk

AI is not a mature technology with stable feature sets and predictable performance curves. The AI systems being sold to law enforcement agencies in 2026 are the product of a technology generation that did not exist five years ago. The systems that will be available in 2031 are not yet built, and the systems available in 2033 will be meaningfully different from what is deployed today. A ten-year contract signed in 2026 expires in 2036. The AI technology landscape of 2036 cannot be predicted from 2026 any more than a procurement officer in 2016 could have predicted the AI capabilities of 2026.

The technology change risk is not merely that better tools will be available. It is that the vendor's own tool may change in ways that are detrimental to the agency. When a vendor updates their AI model, the agency does not necessarily get a better tool. It gets a different tool. The model that produced accurate use-of-force narratives may be replaced by a model with different behavior, different training data, and different gap-fill patterns. If the contract does not include a technology-change provision requiring vendor notification and agency consent before model changes are deployed, the agency may discover the change only when a review of the correction logs shows officers are now correcting different error types than before.

Pricing Risk

A ten-year contract with annual price adjustment clauses is not a fixed-price contract. The typical public-safety technology contract includes an annual adjustment provision pegged to a consumer price index or a stated cap, commonly 5 to 8 percent per year. Compounded over ten years, a contract that begins at $4.5 million per year could reach $7 to $9 million per year by year ten under a modest annual adjustment schedule. The total contract cost in year ten may be 50 to 100 percent higher than year one pricing. The city council that approved a $45 million figure may find that the actual ten-year cost is substantially higher.

Price risk is compounded by vendor acquisition. Public-safety technology vendors are active in the mergers-and-acquisitions market. A vendor acquired by a private equity firm in year four of a ten-year contract may operate under a different pricing philosophy than the vendor the agency evaluated. The contract that looked like a partnership relationship with a company whose leadership the agency knew may become an administrative relationship with a holding company whose primary obligation is to its investors, not to the agency. Contract terms that the original vendor accepted through relationship negotiation may be interpreted differently by an acquiring firm with a mandate to extract value from its portfolio.

Performance Risk

A vendor who holds a ten-year sole-source contract with a law enforcement agency has a different incentive structure than a vendor competing for renewal every three years. The competitive pressure that keeps a vendor's support team responsive, keeps their implementation team engaged, and keeps their product development roadmap aligned with agency needs dissipates once the contract is signed and the renewal is a decade away. This is not a claim of bad faith. It is a structural observation about how competition affects service quality. An agency that evaluates vendor performance every three to five years and can credibly threaten non-renewal has leverage. An agency locked into a ten-year contract has much less leverage in years six through nine.

Performance risk includes the quality of the AI tool's output as well as support response times, implementation quality, and product development. An AI model that performs well in year one of a ten-year deployment may receive less investment in model maintenance and improvement as the vendor shifts development resources toward winning new accounts rather than serving existing ones. The agency's quarterly review cadence, described in the vendor evaluation lesson, is the mechanism for detecting performance decline. Without a contract that includes performance metrics and a termination-for-cause provision tied to those metrics, the review findings have no enforcement mechanism.

Exit Risk

Exit risk is the practical challenge of leaving a sole-vendor contract before the term ends. Early termination provisions in public-safety technology contracts typically require payment of a substantial portion of remaining contract value, often 50 to 75 percent of the unpaid balance, as a termination fee. For a $45 million ten-year contract, early exit in year five could require payment of approximately $15 to $22 million in termination fees on top of the $22.5 million already paid. A city council that approved the original contract for its public-safety value may find it politically and financially impossible to authorize a termination payment of this scale, even if the tool has become genuinely problematic.

Data exit risk is equally significant. The agency's evidence footage, AI-drafted reports, correction logs, audit trails, and case files may be stored in a vendor-managed cloud environment. If the contract ends, or if the agency attempts to exit early, the vendor's obligation to return that data is governed by the contract's data portability provisions. A contract that provides for data return in proprietary format, with a 30-day retrieval window, may leave the agency unable to access its own evidence for ongoing cases after exit. An agency that cannot produce evidence footage because it was stored in a vendor's proprietary cloud system and the contract has ended faces an evidentiary and legal problem that no amount of after-the-fact legal argument can fully resolve.

How to Negotiate Out of the Trap

The ten-year contract trap is not inevitable. Its specific terms can be negotiated, and agencies that understand the risk structure before signing have leverage the agency after signing does not have. The following provisions should be non-negotiable in any long-term public-safety AI contract.

Performance Metrics and Remediation

Every long-term contract for public-safety AI should include specific, measurable performance metrics and a remediation process tied to them. The metrics should cover output accuracy (not the vendor's claimed accuracy rate but a measured rate from the agency's own quarterly review cadence), system availability, support response times, and model stability (notification and consent requirements before model changes are deployed). The remediation process should give the vendor a specific time period to cure a performance failure and give the agency the right to terminate for cause without a termination fee if the vendor fails to cure.

Performance metrics must be set before the contract is signed, not negotiated after a performance failure has occurred. A contract that says the vendor will "maintain acceptable performance levels" gives the agency nothing. A contract that says the vendor will "maintain a minimum 90 percent accuracy rate on patrol narrative drafts, measured quarterly by the agency's review cadence against the agency's own incident set, with a 90-day cure period for any quarter falling below 85 percent" gives the agency an enforcement mechanism. The difference between these two provisions is the difference between a contract that protects the agency and one that protects the vendor.

Policy Change and Force Majeure Provisions

A policy-change termination clause allows the agency to exit the contract without a termination fee if a change in law, prosecutorial policy, or court ruling makes the contracted tool unusable or legally prohibited. This provision should be drafted broadly enough to cover state legislative changes, city council resolutions, changes in CJIS Security Policy that the vendor cannot accommodate, and prosecutor policies that conflict with the vendor's disclosure documentation format. Without this provision, a King County-type prosecutorial policy that bars AI-written reports in year three of a ten-year contract leaves the agency paying for a tool it cannot legally use.

The policy-change clause should be mutual: if a policy change mandates a specific feature the vendor has not built, the vendor should be required to build it within a defined period or face the agency's right to exit. This provision protects both parties and creates an incentive for the vendor to stay current with the evolving regulatory environment, rather than treating regulatory changes as the agency's problem alone.

Technology Refresh and Model Change Consent

The technology-refresh provision governs what happens when the AI model is updated. The agency should require advance notification of any change to the underlying AI model (typically 60 to 90 days is reasonable), a description of how the change affects output quality and behavior, and the agency's right to a re-evaluation period before the new model is deployed in production. The agency should also retain the right to remain on the prior model version for a defined period if the new model fails performance standards during the re-evaluation.

Model change consent provisions are relatively uncommon in standard vendor contracts, but they are achievable in negotiation, particularly for larger contracts. A vendor who argues that model change consent is operationally unworkable is telling you something about how frequently they expect to change the model and how much they prioritize agency stability versus internal development velocity. The agency's interest is in a model that is stable enough to be verified and predictable enough to build a review protocol around. A vendor who updates the model monthly has essentially changed the product the agency evaluated and approved.

Data Portability and Exit Terms

The data portability provision should specify: the format in which data will be returned (a widely used, non-proprietary format such as standard video formats and standard database exports), the timeline for return (not less than 90 days, not subject to vendor discretion), the completeness of the return (not just reports but footage, AI drafts, correction logs, audit trails, and metadata), and the absence of any fee for data return. The vendor's obligation to return data should survive contract termination, including early termination, regardless of the reason for termination.

Exit terms should include a termination fee schedule that decreases as the contract progresses, reaching zero in the final year. A termination fee that equals 75 percent of remaining contract value in year two but decreases to 25 percent in year seven is more equitable than a flat percentage throughout the term. The agency should also negotiate a right to exit without fee if the vendor is acquired by a competitor, because a vendor acquisition changes the material character of the contractual relationship without the agency's consent.

The Alternative Structure: Shorter Terms and Open Competition

The structural alternative to the ten-year bundled contract is a portfolio approach: shorter contract terms, component-level rather than bundle-level contracts, and regular open competition at renewal. This approach sacrifices some of the integration benefits of the bundle and requires more active contract management. It is also substantially more protective of the agency's interests over a decade-long horizon.

A portfolio structure separates the camera hardware contract from the cloud storage contract from the AI software contract. These components have different technology maturity rates, different competitive markets, and different optimal contract lengths. Camera hardware has a longer useful life and a more stable technology curve; a five-year hardware contract is reasonable. Cloud storage is a commoditized market with active competition; a two-to-three-year contract is appropriate. AI software is the fastest-moving component, where the competitive landscape changes most quickly and where a three-year contract cycle allows the agency to evaluate and adopt better tools as they emerge.

Separating the components also reduces the magnitude of any single vendor failure. If the AI software vendor fails to perform and the agency needs to switch, it can replace the software component without also replacing cameras and cloud storage. The data portability provisions for each component can be negotiated independently, reducing the concentration of data risk in a single vendor relationship.

Open competition at renewal is the most effective check on vendor complacency and price escalation. When a vendor knows the agency will issue a full competitive solicitation at the end of each contract term, the vendor's incentive to deliver strong performance throughout the term is aligned with the agency's incentive to pay for strong performance. A vendor who cannot win an open competition at renewal is a vendor whose value proposition has not held up, and the agency is better off knowing that before year ten than after.

Presenting the Contract Risk to City Council

Command staff and chiefs who understand the contract trap face a political communication challenge: how to explain the risk of a long-term sole-vendor deal to a city council that is attracted to the bundle's simplicity and the vendor's upfront discount. The communication task is not to oppose AI procurement. It is to advocate for AI procurement structured to protect the agency over the contract's full term.

The narrative structure that works: acknowledge the operational benefits of the bundle (real and specific), quantify the lock-in risk (use the financial terms: termination fees, price escalation, data exit costs), and present the alternative structure as a way to achieve the operational benefits without the decade-long lock-in. A council that sees the choice as "bundle with lock-in" versus "no bundle and no AI" will usually choose the bundle. A council that sees the choice as "bundle with lock-in" versus "portfolio approach with the same operational benefits and significantly lower risk" can make a more informed decision.

The Brady and Giglio framing is relevant here as well. A council should understand that the AI tool the agency signs a ten-year contract for is not just an operational tool: it is a component of the agency's evidentiary process. A contract that locks the agency into a tool that a future prosecutorial policy change makes unusable, or that a future court ruling makes problematic, is a contract that affects the agency's ability to prosecute cases. That is not an IT procurement question. It is a public-safety mission question, and it is the question a city council should be asked to consider.

Key Takeaways

• Bundled sole-vendor contracts for cameras, drones, cloud storage, and AI software on the order of approximately $45 million and up to ten years are already in the market. The bundle's integration benefits are real, but the dependency structure it creates is the most significant contract risk in public-safety AI procurement today.
• The five specific risks of a decade-long sole-vendor deal are: policy change risk (the governing policy may change before the contract ends), technology change risk (the vendor's AI model may change in ways detrimental to the agency), pricing risk (annual adjustment clauses compound substantially over a ten-year term), performance risk (competitive pressure to deliver dissipates once a long-term contract is signed), and exit risk (early termination fees and data portability gaps can make leaving the contract financially and operationally impossible).
• A policy-change termination clause is non-negotiable. The King County prosecutorial policy barring AI-written police reports illustrates precisely why: a policy change in year three of a ten-year contract can leave an agency paying for a tool it cannot legally use.
• Performance metrics must be specific and measurable before the contract is signed. "Acceptable performance levels" is not a performance metric. A defined accuracy rate, a quarterly review mechanism, and a cure period tied to a termination-for-cause right is a performance metric the agency can enforce.
• Model change consent provisions require the vendor to notify the agency before changing the underlying AI model, give the agency a re-evaluation period, and allow the agency to remain on the prior model version if the new model fails performance standards. These provisions are achievable in negotiation.
• Data portability provisions must specify format, timeline, completeness, and the absence of exit fees. An agency that cannot retrieve its own evidence footage and audit trails after contract termination has a legal and operational problem no amount of after-the-fact argument can resolve.
• The portfolio alternative, separating hardware, cloud storage, and AI software into independently contracted components with shorter terms and open competition at renewal, achieves the operational benefits of integration without the decade-long sole-vendor lock-in.
• Presenting the contract risk to city council requires the narrative to be framed as an AI procurement structure question, not an anti-AI position. The council is more likely to act on contract risk when it is presented as a way to protect the public-safety mission over the contract's full term, including the risk that a future policy or legal change makes the contracted tool unusable.