As nations prepare for a critical UN negotiating session on lethal autonomous weapon systems (LAWS), a comprehensive new blueprint lays out exactly what's at stake: the question of whether humans will retain meaningful control over life-and-death decisions made by AI-powered weapons. The document, prepared by researcher Cian Westmoreland for the Global Governance Institute, catalogs 121 distinct risks across technological, military, legal, political, ethical, and societal domains, then maps out specific, operationally grounded solutions that delegations can actually achieve in the final negotiating window before the November 2026 Review Conference.

The urgency is real. Autonomous weapons systems are no longer theoretical. Real-world cases like Lavender (used in Gaza), Lancet (deployed in Ukraine), Maven (a U.S. military AI system), and Kargu-2 (a Turkish drone system) show that the technology is already operational, however contested the details of individual deployments may be. The central problem, according to the blueprint, is that current regulatory frameworks have a critical gap: they focus on fully autonomous weapons but miss the decision-support systems and information systems where algorithmic influence over targeting is most pervasive.

What's the Core Problem With Current Autonomous Weapon Rules?

The blueprint diagnoses what it calls "The Kill Cloud," a socio-technical system combining orbital sensors, commercial AI, autonomous weapons, and distributed command infrastructure. The engineering objectives were speed, resilience, and deniability. But the structural consequence is that human moral and legal agency has been architecturally displaced from lethal force decisions. Existing treaties and regulations don't adequately address this reality.

The rolling text under negotiation contains language that formally excludes decision-support systems from the instrument's scope. That's a problem because it leaves a regulatory blind spot precisely where algorithmic influence is most dangerous. The blueprint argues that silence on this issue is preferable to formal exclusion, because silence can be addressed in future protocols, whereas formal exclusion cannot be easily reversed.

How Can Negotiators Make Progress in the Next Few Weeks?

Rather than proposing an all-or-nothing overhaul, the blueprint identifies three concrete textual interventions that are achievable in a single negotiating session:

• Decision-Support Systems Non-Exclusion: The rolling text should not formally lock in decision-support systems exclusion from the instrument's scope, preserving the possibility of future regulation.
• Meaningful Control Operationalization: The phrase "context-appropriate human control and judgment" should be conditioned on three testable criteria: epistemic agency (the operator must know enough to make a real decision), deliberative agency (there must be time and cognitive bandwidth available), and causal agency (human input must materially shape the outcome, not merely endorse a precomputed recommendation).
• Information System Environment Testing: Obligations and testing requirements should explicitly cover the information system environment in which identification, selection, and engagement functions operate, including adversarial degradation scenarios like GPS denial and sensor spoofing.

The blueprint also proposes a tiered instrument structure that gives delegations multiple landing zones for negotiation. A binding prohibition on fully autonomous anti-personnel use would be the floor. A strong transparency baseline, including information system disclosure requirements, would be mandatory. An optional full audit trail framework (using distributed ledger technology) would be available for states willing to accept it. This structure avoids all-or-nothing trade-offs and creates a viable negotiating shape for the August session.

What Does "Meaningful Human Control" Actually Mean in Practice?

One of the blueprint's key contributions is operationalizing the concept of meaningful human control, which has been debated for years without clear definition. The framework proposes that any treaty definition rest on three operationally testable criteria. First, epistemic agency: the operator must have access to enough information to make a genuine decision, including confidence scores, data sources, and failure modes at the point of use. Second, deliberative agency: there must be adequate time, cognitive bandwidth, and access to counter-evidence. Third, causal agency: the human input must materially shape the outcome, not merely rubber-stamp a precomputed recommendation.

The blueprint acknowledges that these proposals are not costless. They impose latency on engagement decisions that may be operationally unwelcome. They require infrastructure investment that will fall most heavily on states with limited resources. They raise classification and interoperability questions for coalition operations. But the blueprint argues that the costs of unconstrained, unaccountable autonomous weapon deployment are substantially higher, more durable, and harder to reverse.

What Role Does AI Compliance Play in Broader Governance?

While autonomous weapons represent one frontier of AI governance, the broader regulatory landscape is tightening across commercial AI systems as well. The European Union's AI Act is being enforced on a phased timeline, with prohibited practices already banned and transparency obligations for limited-risk systems in effect. In the United States, the Federal Trade Commission (FTC) is accelerating enforcement of AI-related deceptive and unfair practices, and state-level regulations are closing gaps that federal law has not yet addressed.

Organizations deploying large language models (LLMs), which are AI systems trained on vast amounts of text data to generate human-like responses, and other AI-powered applications face real financial consequences for non-compliance. The EU AI Act establishes penalties based on violation severity: prohibited AI practices can trigger fines up to 35 million euros or 7% of an organization's global annual turnover; violations of high-risk and transparency obligations carry penalties up to 15 million euros or 3% of global annual turnover.

Two 2024 cases illustrate the practical stakes. Air Canada was held liable after its AI-powered chatbot gave a passenger fabricated information about a bereavement discount policy. The British Columbia Civil Resolution Tribunal ruled that Air Canada had to honor the policy the chatbot made up and pay the passenger for fees and damages. In another incident, security researchers identified a vulnerability in Salesforce's Slack platform that would allow attackers to use an AI-powered feature to steal information from private channels through prompt injection, a technique where attackers manipulate the language model to behave unexpectedly. Salesforce patched the vulnerability within days.

What Are the Core Pillars of AI Compliance Programs?

Organizations building compliance programs face a complex landscape of requirements. The core pillars of an audit-ready AI compliance program include:

• Data Privacy and Protection: Ensuring that personal data used in AI systems is handled according to applicable regulations like GDPR and HIPAA.
• Transparency and Disclosure: Providing clear information to users and regulators about how AI systems work and what data they use.
• Fairness Controls: Implementing mechanisms to prevent discriminatory outcomes in AI decision-making.
• Security and Supply Chain Integrity: Securing AI systems against cyberattacks and ensuring that open-source components and dependencies are vetted and tracked.
• Human Oversight and Accountability: Maintaining meaningful human control and clear responsibility for AI system decisions.
• End-to-End Documentation: Creating full traceability records across the entire AI lifecycle, from training data sourcing through deployment and monitoring.

Many organizations overlook a growing source of risk: the open-source packages, models, and containers that comprise the AI supply chain. Research from Omdia found that of 500 IT leaders surveyed across North America and the UK, only 17% have a unified observability strategy spanning teams, infrastructure, and governance. This means most organizations cannot trace a decision back through the environment, dependencies, and data that produced it, which is precisely what regulators now require.

The EU AI Act's technical documentation requirements, specifically Articles 11 and 17, require high-risk AI system providers to document data sources, component dependencies, provenance records, and validation evidence. Training data sourcing and validation of third-party components both fall within this requirement, as do controls over sensitive data handling. Many organizations lack sufficient data governance practices at this level of the technology stack and cannot produce required documentation on demand.

The convergence of these two governance frontiers, autonomous weapons and commercial AI compliance, reflects a broader shift: AI governance is moving from abstract principle to concrete, enforceable requirement. Negotiators preparing for the August 2026 UN session on autonomous weapons face the same fundamental challenge that enterprises face with compliance programs: translating high-level principles like "meaningful human control" and "transparency" into operationally testable, auditable standards that can actually be implemented and verified.