This weekly update brings major news about keeping AI agents safe and protecting company data. Several big companies made important announcements about agentic AI security.

Proofpoint, a cybersecurity company, announced four new ways to protect workplaces where people and AI agents work together. They call this the "agentic workspace." The company warns that AI assistants can be tricked by bad actors through email attacks. These attacks use prompt injections to make AI agents share secret information or give wrong answers to users.

Experts are worried because companies are adding AI agents to their work very quickly. A new report shows that by the end of this year, there will be over 45 billion non-human identities working in companies. This includes AI agents that can make decisions and do tasks by themselves. However, only 10% of business leaders have a good plan to manage these AI workers safely.

The problem is that AI agents need access to company data to do their jobs well. But this also makes them dangerous if hackers find ways to control them. Bad actors can use prompt injection attacks to trick AI agents into sharing private information or doing harmful actions. Since 80% of security breaches involve stolen identities, protecting AI agent identities is very important.

Splunk, another security company, added agentic AI to help security teams work faster. Meanwhile, Intuit improved their AI system called GenOS to build better AI agents for financial tasks. They created special Financial AI models that work 50% faster and 5% more accurate than regular AI models for accounting work.

Companies need to be very careful when giving AI agents access to sensitive data. Security experts recommend using granular access policies and making sure AI agents only get the permissions they need for limited times.