Security | Threat Detection | Cyberattacks | DevSecOps | Compliance

Imagine getting three separate security reports back for your new enterprise AI assistant: On paper, the application looks ready for production, but in reality, a threat actor bypasses your guardrails in minutes. How? By using the AI model as an intermediary. The attacker steers the LLM to invoke the internal utility tool, thereby bridging an untrusted prompt directly to the backend execution sink.

Most organisations now run at least one LLM in production, and a growing number run agents that call tools and act without a human in the loop. The security testing those systems receive was designed for deterministic software. AI applications fail differently. The payload is natural language, the same input can be safe nine times and unsafe on the tenth, and the malicious instruction often arrives inside a document or tool description rather than from the user.

Containment guidance for agent incidents already exists and it is largely correct. Revoke the tokens, freeze the orchestration tier, cut egress, set the vector store to read-only. Those steps take minutes and any competent team will find them. ‍ The difficulty sits either side of containment. Deciding what kind of incident this is takes longer than stopping it, establishing what the agent did before you stopped it takes longer still, and both depend on preparation that has to exist beforehand.

In July 2026, two AI stories broke that appeared unrelated on the surface. But were they really? The first was an AI product's shared conversation links, meant for specific people, turning up in Google searches, some holding sensitive personal and company data. The second was a frontier AI lab's own model escaping a security sandbox during an internal evaluation and spending four and a half days inside three companies' systems. One involved ordinary users making a common mistake.

AI coding assistants have gotten remarkably good at writing functional code. Syntax correctness rates are approaching 100%. Developers are more productive than ever. And yet the security picture tells a very different story.

Zombie APIs are API versions or endpoints that were once known and documented, but were never properly retired. A team ships v2 of an API, tells everyone to migrate, and assumes v1 is dead. In reality, v1 is still running on a server somewhere, still accepting requests, and still connected to production data. This is different from unmanaged APIs, which were never documented in the first place. Zombie APIs were documented once.

Aikido Security has achieved ISO 42001:2023 certification, the international standard for AI management systems, a step few security vendors have taken so far. The certification confirms that Aikido runs a structured, continuously improving governance system for managing the risks introduced by its AI-enabled features, across our entire platform.

TL;DR: Aikido now pentests Android apps. The same agents that test your web apps and APIs can now work through your APK, log into the app, and reason through it, alongside the backend it talks to, in a single assessment. Findings come back with reproduction steps and are ready for an AutoFix, the same as any other Aikido pentest. Aikido has been running autonomous pentests against web apps and APIs since November 2025.

Healthcare has always run on two currencies: information and trust. Patients hand over their most sensitive data, diagnoses, genetic profiles, mental health histories, on the assumption that it will be protected and that the people (or systems) handling it will treat them with care. Artificial intelligence is now reshaping both sides of that bargain at once.

Along with the Bank of England in May, the ECB is requiring organisations to produce an action plan against AI-driven cyberattacks (opens in a new tab). What it wants, however, is not a strategy paper but proof of operational readiness.