March 2026 Summaries

AI security has evolved beyond traditional approaches as AI systems now permeate various facets of organizational operations, such as employee workflows, application functionalities, and autonomous agent actions. The nature of AI risk is inherently fragmented, with threats stemming from human interactions, probabilistic outputs, and system actions, which makes it challenging to address these risks in isolation. Organizations typically attempt to tackle AI security through disparate solutions, but the interconnectedness of AI systems demands a unified approach, known as the AI Defense Plane. This approach consolidates visibility, protection, and governance across all AI interactions, enabling organizations to track risk across the entire lifecycle of AI from input to action. As AI transitions from generating outputs to executing actions, the repercussions of security lapses increase, necessitating a coordinated security strategy that views AI as an integrated system rather than a collection of isolated tools.

The Backbone Breaker Benchmark (b3) is developed by Lakera in partnership with the UK AI Security Institute to assess the security of backbone large language models (LLMs) against adversarial attacks. Unlike evaluations focused on capability or safety, b3 isolates the core model powering AI agents to test its resilience against manipulation. Using data from nearly 200,000 human red-team attacks collected via the Gandalf: Agent Breaker challenge, the benchmark creates structured "threat snapshots" that simulate real-world attack scenarios. These snapshots evaluate model responses at various defense levels to determine vulnerability. The benchmark employs different scoring methods according to attack objectives, offering insights into how effectively models resist manipulation. Researchers can run the benchmark using tools from the Inspect Evals GitHub repository, and the evolving nature of b3 aims to keep pace with advancements in AI and emerging attack techniques, contributing to a shared empirical approach for measuring AI security.

AI Gateways serve as a centralized control layer that standardizes access to large language models (LLMs) and agent systems across various providers, facilitating scalable enterprise AI deployment. These gateways offer a range of functionalities, including enforcing guardrails and compliance, managing costs, providing telemetry and observability, and abstracting provider interactions, which help organizations avoid the fragmentation and governance sprawl that often accompany unmanaged AI adoption. Positioned between applications and AI providers, an AI Gateway ensures consistent policy enforcement and governance without replacing the AI systems themselves, making it essential for organizations with complex, multi-provider AI setups. By shifting governance from application logic to infrastructure, they allow teams to focus on developing AI applications without the burden of implementing individual governance measures, thereby accelerating onboarding and reducing friction. While not necessary for all organizations, AI Gateways become indispensable as AI complexity and provider diversity increase, transforming AI governance into an organizational policy rather than a project-level concern.