Different systems. Same control point.

Agents handle rebooking, refunds, fee waivers, and disruption recovery against the passenger service system, where one wrong action becomes a binding commitment to the passenger and a cascading operational cost. IntentGate authorises every call before it reaches Amadeus, Sabre, or Navitaire, holds irreversible actions above a value threshold for a named approver, blocks unbounded operations so no single agent can mass-refund or mass-rebook, checks the action against the task the agent was actually given so an instruction planted in a passenger message cannot waive a fee, and writes a tamper-evident record of every decision for Part-IS and NIS2 oversight.

Agents handle payments, disputes, onboarding, and adviser support, where a single wrong action moves money, breaks an AML or PCI-DSS control, or mishandles regulated data. IntentGate authorises every call, holds transfers and high-value actions for approval, keeps automated loan and claims agents inside policy limits, and writes a tamper-evident record that supplies the DORA incident fields directly from the audit log.

Insurers run agents across claims triage, underwriting, and fraud checks, where an autonomous decision pays out money or accepts a risk. The exposure is an agent approving a claim or binding cover it should not, a manipulated input steering an underwriting decision, or regulated policyholder data leaving through an agent. IntentGate authorises each action against the declared task, holds claim payouts and underwriting approvals inside policy limits for human sign-off, filters restricted data on the way out, and keeps a tamper-evident record for Solvency II and GDPR accountability.

Agents touch the electronic health record (EHR), scheduling, and clinical databases, where GDPR special-category data and patient safety are both in play. IntentGate scans and filters outgoing agent responses so protected health data cannot leave the secure environment, keeps GDPR and EHDS (European Health Data Space) minimum-necessary obligations intact, and enforces a mandatory human-in-the-loop hold on high-impact actions so no manipulated agent can act unsupervised, satisfying the EU AI Act human-oversight mandate.

Drugmakers deploy agents to accelerate molecular design, manage clinical trials, and run automated manufacturing under GxP. A drifting or compromised agent risks leaking patented molecular IP to public LLMs, corrupting clinical-trial records or patient eligibility, or altering sterile production parameters. IntentGate scans outgoing agent payloads against your configured patterns to block proprietary molecular structures and genetic sequences from leaving the network, requires verified human-in-the-loop approval before an agent can modify a trial database or a production parameter, and generates the secure, time-stamped, hash-chained audit trail that FDA 21 CFR Part 11 record-keeping and GxP audits call for.

Agents in citizen services and case handling make decisions about people and touch sensitive data that must remain strictly inside a sovereign cloud boundary, such as SecNumCloud or FedRAMP, to maintain full accountability. Because IntentGate is self-hosted, it runs entirely inside that sovereign environment, preventing unauthorised data egress, records every autonomous decision alongside the exact policy that produced it, holds high-impact actions for human authorisation, and surfaces each agent for mandatory EU AI Act classification.

Grid operators and water and wastewater utilities run agents to optimise flow rates, chemical dosing, and plant operations across IT and the SCADA systems that control the physical grid and valves. The exposure is an optimisation or hijacked agent that oversteps, shedding load that a hospital depends on, or pushing a treatment valve past a safety limit or outside EPA-approved chemical thresholds, alongside the Langflow flaw (CVE-2025-3248) that put exposed agent-framework servers on CISA KEV after real-world compromise. IntentGate uses signed capability scope to block optimisation agents from reaching SCADA or OT controllers, holds any command that violates a physical safety limit for verified human sign-off, blocks self-escalation, and contains an anomalous agent with a per-agent kill switch and a hash-chained audit that meets the NIS2 reporting clock.

Upstream and midstream operators put agents across drilling rigs, high-pressure pipelines, LNG terminals, and parts procurement, and some trading desks let agents act on live market data. The exposure is a hijacked or hallucinating agent pushing a pump or refinery valve past its safety limit, a confused-deputy loop placing millions in duplicate non-refundable emergency orders, or a trading agent fed spoofed data to dump or overbuy reserves. IntentGate enforces default-deny between logistics and physical OT, holds any pressure or flow-rate change and any high-value trade or order for verified human sign-off, and isolates a compromised pipeline agent instantly with a per-agent kill switch before it can move laterally.

Carriers, terminals, and port operators run agents across administrative IT (customs filing, freight documentation, spot-rate quotation) and operational technology, interfacing with the Terminal Operating System (TOS, such as Navis N4) and Port Community System (PCS) that drive crane automation and automated guided vehicles. A compromised agent could issue or amend a bill of lading, which is a document of title, file a false customs declaration, misdeclare hazardous cargo, evade sanctions, re-route thousands of containers into demurrage, collapse spot rates on a trade lane, or bridge from IT into physical OT. IntentGate gives each agent a narrow, signed capability scope, so a customs or booking agent can never reach crane-control or TOS APIs, holds irreversible and high-value actions for a named approver, blocks unbounded operations so no single agent re-routes a whole service at once, keeps agent-to-agent traffic default-deny, provides a per-agent kill switch, and produces the tamper-evident audit that supports IMO cyber risk management and the NIS2 duties ports carry as essential entities.

Shipyards and naval builders run agents across design, procurement, and yard operations, where IT and OT sit close on flat networks with satellite-edge links. The exposure is agentic automation hijacking, lateral movement from a compromised terminal toward engine or steering controls, and social engineering that tricks staff into letting an agent act. IntentGate keeps agent-to-agent traffic default-deny and segmented, holds high-impact and irreversible actions for a human, provides a per-agent kill switch to contain a compromise, and records every decision.

Makers of aircraft, spacecraft, and defence systems put agents into design and engineering (CAD, PLM, specifications), procurement, and production across highly integrated networks. The exposure is proprietary or classified design leaking to public AI, ITAR-controlled or model IP pulled out through an agent, and an agent being steered into an action it should not take. IntentGate authorises each action against the declared task, holds high-value and irreversible actions for approval, filters restricted design data on the way out, and, because it is self-hosted and in-tenant, keeps ITAR and export-controlled data in-tenant with the tamper-evident audit that CMMC and NIST 800-171 expect.

Agents now sit in procurement, supply-chain planning, and connected-product operations, reaching the ERP and PLM systems where the supplier master, source code, and product specifications live. IntentGate authorises each action against the declared task, holds high-value or irreversible orders, and filters restricted data on the way out, so an over-scoped agent cannot place a rogue order or leak the supplier master.

Automakers and OEMs are turning cars into computers on wheels, with agents reaching connected-vehicle commands, over-the-air updates, dealer financing, and warranty and recall handling. A manipulated agent could send a command toward a vehicle or its controls, approve financing it should not, or leak driver data. IntentGate authorises each action against the declared task, holds high-impact vehicle and financial actions for a human, filters driver data on the way out, and keeps a tamper-evident record for UNECE R155/R156 and EU AI Act auditability.

Agents drive autonomous ordering, warehouse robotics, shipping routing, and distribution across the WMS and TMS (warehouse and transportation management systems), where one misread signal cascades. A rogue agent could place multi-million-euro panic-orders after misreading a temporary shipping delay, or a confused-deputy chain could quietly act at scale. IntentGate authorises each action against the declared task and its limits, holds high-value orders for approval, and keeps east-west traffic segmented, so a poisoned or misread instruction cannot cascade.

Construction and engineering firms deploy agents to schedule projects, coordinate machinery logistics through the ERP, and parse CAD and BIM plans and contracts. The exposure is a prompt injection subtly altering load tolerances or specifications, a panic-order for millions in materials or a leak of subcontractor bids, or a missed notice-of-delay that waives a contractual right. IntentGate holds writes to CAD and BIM data for dual-engineer sign-off, flags and holds abnormal high-value material and equipment orders, enforces egress boundaries so bids and pricing cannot leak, and records every action.

Chemical producers deploy agents to monitor reactor conditions, manage safety data sheets, and automate laboratory workflows across IT and OT. Because they handle high-hazard substances, a hijacked or drifting agent could push temperature, pressure, or concentration past safe limits or release a toxic formulation. IntentGate uses signed capability scope to block lab and optimisation agents from reaching plant SCADA, PLC, and safety systems, freezes any action that exceeds a safety threshold for human sign-off, and creates audit-ready logs for REACH and related chemical regulations.

Food and beverage manufacturers use agents to optimise ingredient mixing, manage batch records, and handle supply traceability. An ungated agent is exposed to shadow use and prompt exploits that can cause a recipe deviation, an allergen error, or a failed food-safety audit. IntentGate discovers and contains unsanctioned agents touching batch and supply records, holds any command that deviates from an approved recipe or allergen standard, and keeps verifiable traceability records for EU and FDA reporting under tight audit deadlines.

Farm operators and agribusinesses deploy agents to dispatch autonomous tractors and harvesters, regulate automated irrigation, and optimise chemical spraying, while agri-tech researchers use assistants over cloud-hosted seed and GMO databases. Because these agents bridge digital commands with physical field operations, a compromised or manipulated agent risks forcing machinery off safe paths, over-applying pesticide or fertiliser past eco-safety limits, or leaking patented seed genetics. IntentGate cryptographically isolates farm-management agents from sending physical commands to tractors or spray valves without a hard gate, enforces limits on dosing and water-routing tools so an agent cannot exceed eco-safety parameters, and screens outgoing responses so proprietary crop genetics and farm telemetry stay private.

Operators moved fast from support chatbots to agents that reach into the OSS and BSS, the operations and business support systems that provision SIMs and eSIMs, change plans, and apply billing credits. Every one of those is a real action with money or an identity behind it. IntentGate holds high-impact OSS/BSS actions for verified human approval, so a manipulated agent cannot complete an unauthorised SIM swap or a free-plan billing override.

Law firms, consultancies, and accounting firms deploy agents to draft contracts, review M&A documents, analyse tax structures, and file with court, patent, and regulatory portals. A drifting or compromised agent risks leaking attorney-client privileged data to an external LLM, bypassing an ethical wall to reach a conflicting client's files, or autonomously submitting a flawed or hallucinated filing. IntentGate redacts or blocks privileged data and trade secrets in outbound payloads, treats any court, patent, or SEC submission as a high-impact action a licensed professional must sign off, and binds each agent's retrieval scope to the active matter so it cannot cross an ethical wall.

Agents run pricing, refunds, and customer service, the exact attack surface where a crafted prompt can trigger a €1 checkout or an unwarranted refund at the payment gateway. IntentGate binds each action to the declared task and validates it against security policy, PCI-DSS scope, and transaction limits, so injected instructions are treated as untrusted data, never as commands.

Ride-hailing, lodging, delivery, and e-commerce platforms deploy agents to resolve disputes, onboard hosts and merchants, and coordinate dynamic pricing. Because they handle massive transaction volumes, they are highly vulnerable to prompt-injection exploits designed to hijack account payouts, trigger fraudulent refunds, or manipulate marketplace rates, and in a multi-step workflow a hijacked hop can abuse a generic service credential to reach restricted data downstream. IntentGate cryptographically binds every tool call to both the acting agent and the initiating user's validated identity across every hop, intercepts and holds high-impact financial actions such as large payouts, refunds, and fee adjustments for operator clearance, and treats all natural-language inputs and listings as untrusted data, never as commands.

Media, gaming, and entertainment companies put agents into content operations, ad buying, moderation, and in-game economies. The exposure is an agent moving ad budget or issuing in-game currency, refunds, or payouts beyond policy, a manipulated prompt publishing or unpublishing content, or user data leaving through an agent. IntentGate binds each action to the declared task, holds high-value spend, payouts, and publishing actions for approval, and keeps a tamper-evident record of every decision.

Universities, schools, and EdTech platforms deploy agents to grade work, run personalised tutoring, and screen admissions and scholarships, decisions that directly affect people. The exposure is an autonomous screening agent discriminating silently at scale, protected student records sent to an unapproved public LLM to answer a query, or a portal prompt injection changing a grade or firing an automated suspension. IntentGate holds admissions, grading, disciplinary, and scholarship actions for human review rather than letting an agent decide alone, redacts student records on the way out to keep GDPR and FERPA intact, and keeps a tamper-evident record of every decision.