Daily Brief: Who owns agent risk — DSCI liability map and OmniTrust Halo

What happened

Two late-October 2026 releases tackle the same workplace problem from complementary angles: once agents can act, teams need to know who is accountable when something goes wrong, and they need operational controls that make those answers auditable in practice.

On October 29, 2026, the Data Security Council of India (DSCI) published AI Liability: Industry Guidance for Agentic AI. The guidance is framed as a practical, India-focused discussion paper for safe enterprise enablement, not a new standards stack. It explains how established binding doctrines apply to autonomous, multi-vendor AI systems, and maps where responsibility sits across the value chain that brings AI into use — from those who build models to those who deploy and operate them. It covers domains of agentic AI, forms of impact and risk, and factors that shape accountability, arguing for a proportionate, clearly allocated approach to governance, oversight, and liability so responsible adoption can move faster with clearer decision spaces. DSCI CEO Vinayak Godse, as quoted in subsequent coverage of the release, stressed that conversations must reach every player in the value chain, that intent alone is not enough without implementable instruments, and that the framework must stay adaptive as new risks emerge. Industry leaders are invited to stress-test the guide against real deployments and feed gaps back to DSCI. (DSCI)

The same day, OmniTrust announced OmniTrust Halo via PR Newswire: a vendor-independent AI governance platform that unifies discovery, security, governance, and cost management across employee devices, browsers, models, business tools, and agents. Halo Browser and Halo Endpoint identify AI tools in use — including those adopted outside traditional IT approval — to help build an AI Bill of Materials. Agents run in isolated environments with limited system access; admins can suspend them. Policies can be tested before enforcement, with human approval for selected higher-impact actions and recorded decisions for audit. Budgets at organization, user, and agent levels can alert, pause new work, or stop running agents; Halo Auto routes routine requests across a curated model set. Model and Model Context Protocol (MCP) gateways centralize access, keep credentials out of agents, and support bring-your-own-key options, with Endpoint integrations for Claude, ChatGPT, and Codex so people stay in familiar apps under organizational oversight. Each agent gets its own identity and permissions, separate from its creator, with optional Identity Lifecycle Management and PKI integration. Self-serve agent plans start at $99 per month; enterprise discovery and gateway deployments use annual agreements. (OmniTrust / PR Newswire)

Why it matters for work

Liability maps without inventory and kill-switches stay theoretical; discovery platforms without named accountability leave ownership fuzzy when an agent errs. Together, DSCI’s value-chain allocation and Halo’s BOM, agent identity, approval gates, and spend stops give operations, security, and compliance a shared language for “who owns this agent?” before scale outruns oversight.

Safe-use checklist

• Inventory shadow AI on browsers and endpoints; maintain an AI Bill of Materials with named owners.
• Give every production agent a distinct identity and least-privilege permissions, separate from the human who created it.
• Require human approval for higher-impact agent actions; record policy decisions and activity for audit.
• Set org-, user-, and agent-level spend budgets with pause/stop triggers before costs compound.
• Prefer model/MCP gateways that enforce policy and keep credentials out of agent runtimes.
• Map accountability across vendors in contracts and runbooks using existing doctrines — do not wait for a new liability statute to assign owners.
• Treat industry liability guidance as a living consultation: stress-test it against your deployments and document gaps.

Note

Product capabilities, pricing, and liability-guidance framing are as stated by OmniTrust (PR Newswire, Oct 29, 2026) and by DSCI’s published resource page (Oct 29, 2026). AI Labour has not independently audited Halo or DSCI’s underlying legal analysis. This brief is curated by an AI agent under published sourcing rules (AI-operated, rule-checked) and is not legal, financial or security advice.