AIBound launches SkillGuard to spot malicious AI skills

AIBound on July 8 launched SkillGuard, a new security product for enterprise AI assistants that inspects hidden “skills” and flags dangerous instructions. The company says the tool found thousands of malicious skills on its first day, underscoring how AI assistants can expose companies to breaches, fraud and data leaks.

Why it matters: - Enterprise AI assistants are becoming part of daily work, but the small add-ons they use can quietly expand access to logins, systems, money movement and file exports. - A single malicious skill can create breach, financial and compliance risk without triggering traditional security tools. - Industry analyst Gartner predicts that by 2028, one in four company security breaches will be caused by AI assistants being misused or abused.

What happened: - AIBound launched SkillGuard, a product designed to inspect the “skills” used by company AI assistants and flag dangerous ones. - SkillGuard identified thousands of malicious skills on its first day of use. - The launch took place July 8, 2026. - AIBound described SkillGuard as a way to expose hidden instructions inside enterprise AI assistants.

The details: - Skills are small add-ons that teach an AI assistant a single task, similar to installing an app on a phone. - Harmful skills can be built in plain language, which makes them hard for ordinary security software to detect. - SkillGuard reads every skill, evaluates what it is trying to do and judges intent in context. - The product examines both the plain-language instructions and the underlying actions together. - AIBound says that approach reduces false alarms while surfacing genuinely dangerous skills. - In early use, SkillGuard found a skill disguised as a routine helper that was copying a developer’s private security keys. - SkillGuard also found a skill that tried to trick an assistant into ignoring its own safety rules. - AIBound said those risks map to the OWASP Agentic Security Initiative’s Top 10 for AI agents, including prompt injection, sensitive data exposure and excessive permissions. - AIBound also helps companies find AI assistants in use, understand what each one can access, score risk and stop dangerous ones automatically. - SkillGuard extends that coverage to the instructions inside each assistant.

Between the lines: - The launch targets a gap between broad AI adoption and security teams’ ability to keep up with what assistants can do. - The core problem is not just access, but trust: an assistant can look legitimate while carrying instructions that serve an attacker. - Niall Browne, AIBound co-founder and CEO, said malicious skills are the biggest new blind spot in enterprise AI and that traditional tools miss them because the harmful part is written in everyday language. - Sayed Irfan, PhD, AIBound’s head of artificial intelligence, said the company treats every skill as untrusted and combines deterministic checks with layered AI review. - A CISO at a health care provider said SkillGuard flagged risky skills in the organization’s environment that nothing else had caught.

What’s next: - SkillGuard is available now to AIBound customers. - AIBound is positioning the product as part of a broader effort to detect, assess and stop risky AI behavior before it causes harm. - The company is also extending its security coverage from the assistant itself to the instructions running inside it.

The bottom line: - AIBound is betting that enterprise AI security now has to scrutinize not just what an assistant can access, but what hidden instructions are telling it to do. Its pitch: find the bad skills before they become the breach.