Enterprise AI Agents Pose 'Ultimate Insider Threat'
As enterprise AI agents gain autonomy to launch other agents, spend money, and modify systems, they are creating a new class of security risk that blurs the line with insider threats. In response, Veea Inc. has open-sourced 'Lobster Trap,' an inline watchdog that inspects agent prompts and enforces security policies in sub-milliseconds, highlighting a growing demand for robust governance over agentic workflows.
The open-sourcing of Lobster Trap by Veea Inc. at Mobile World Congress 2026 arrives as enterprises grapple with AI agents that can act independently, creating novel security challenges. Unlike traditional software, these agents can be manipulated through "prompt injection" to leak data or trigger unintended actions, a risk that existing security tools weren't designed to handle. Lobster Trap, written in Go and distributed under an MIT license, operates inline to inspect these conversations with sub-millisecond latency, blocking or logging policy violations. The partnership with NativelyAI aims to embed this security tool directly into the workflows of its 250,000-strong developer community, promoting a "secure by default" approach to agent deployment. This addresses a critical gap, as research shows that while over 90% of AI-driven workflows will involve autonomous agents by 2026, only 21% of organizations have mature governance models in place. The problem is compounded when multiple agents interact, creating complex risks of impersonation and unauthorized escalation of privileges between automated systems. For engineering leaders, the rise of agentic AI elevates the importance of technical due diligence beyond typical M&A checklists. A CTO must now assess not just code quality and infrastructure scalability, but also the architectural constraints that make malicious agent behavior impossible, as written policies alone are insufficient for governing autonomous systems. This requires a shift from observing outputs to validating the fundamental design of how agents can interact with enterprise systems and data. In the adtech space, this trend toward AI-driven automation coincides with the protracted shift away from third-party cookies. After years of delays and industry pushback citing revenue drops of up to 30% for publishers in testing, Google's Privacy Sandbox initiative was formally discontinued in late 2025, leaving cookies in place for now. This leaves the industry focused on first-party data strategies and alternative identifiers, with AI now central to optimizing ad buying, contextual targeting, and managing campaigns across a growing number of channels like CTV and retail media. The UK's tech startup scene is a fertile ground for these trends, with AI-focused companies attracting significant investment. In the first two months of 2026, UK startups raised $4.26 billion, an increase from the previous year. A notable recent funding round was London-based Wayve's £888 million Series D to advance its autonomous driving technology, backed by investors including SoftBank, Microsoft, and NVIDIA. The 2026 Formula 1 season is set for a major reset, kicking off in Melbourne with sweeping new regulations. The changes include smaller, lighter cars, a near 50/50 power split between combustion and electric, and the replacement of DRS with a manual "Overtake Mode" boost, fundamentally altering race strategy and energy management. Pre-season testing saw strong pace from Ferrari, who are experimenting with a novel inverted rear wing design, and Mercedes, while reigning champion Lando Norris begins his title defense for McLaren.
