Meta used its Connect 2026 announcements to expand Muse, its personal AI agent, into a broader shopping and payments ecosystem. The company says it is adding retailers including Walmart, Best Buy, Gap, Sephora and Wayfair, alongside PayPal and Shop Pay for payments.

Muse is also being extended to Meta's AI glasses, giving users a hands-free route to the same agent. Meta says Muse asks for approval before certain consequential actions such as purchases and maintains an audit trail. Its existing design also uses a secure credential store and one-time card numbers for shopping. Those security and privacy claims are Meta's description of its own product and will need independent testing as usage scales.

The commercial significance is the consolidation of discovery, recommendation and transaction execution inside one agent. That changes where influence and responsibility sit. A consumer may no longer move visibly between search, merchant and payment interfaces, while the agent mediates all three. Payments firms and regulators will need to know when the agent is recommending, when it is acting, which party is responsible for errors and how users can unwind a transaction. Agentic commerce is moving from prototype to distribution problem.

Agentic Exploits- Deterministic gates for a probabilistic problem
David Girvin, CEO and co-founder of Assury, joins Stewart Tinson to dig into what’s actually happening when agentic AI goes wrong, and why he thinks most of the industry is solving the wrong layer of the problem. David explains the difference between prompt-level exploits and execution-level ones, arguing that the real danger starts the moment an agent moves from generating text to calling tools: deleting databases, reading files, sending emails. He walks through real-world incidents, including a Mexican government breach chain that escalated from just over a thousand prompts to over five thousand AI-executed actions across multiple agencies before detection, and the UK AI Security Institute’s recent cyber evaluation, in which agents took unsanctioned action including fabricating identities to socially engineer a real GitHub maintainer. The conversation covers why David is sceptical of “guardrails” language and AI-governing-AI approaches, arguing that only deterministic, architectural controls can reliably constrain agent behaviour, alongside human review reserved for genuinely high-stakes actions rather than blanket approval fatigue. He breaks down credential starvation, session risk accumulation, and why classifier-based tools keep failing inconsistently on identical actions, pointing to a named frontier lab’s own zero trust paper as an example of the industry misjudging what actually works. Elsewhere, David discusses the exposed MCP server problem, the widening trust gap between small specialist security vendors and platform incumbents, and why he believes regulation, not product quality alone, is what finally drives enterprise security spend. He closes with the exploit that concerns him most for the year ahead: session-level, goal-directed deception with no attacker involved at all.
BrightTALK

Share this post
The link has been copied!