Call us now:
Massachusetts Institute of Technology evaluated several AI agents and highlighted very serious concerns about their development, documentation, and management.
https://lnkd.in/eDGUwuTk
In its report The 2025 AI Index: Documenting Sociotechnical Features of Deployed Agentic AI Systems, MIT highlighted cases where rogue agents had no documented way of being shut down.
It all sounds very funny and reminiscent of an 1980s film; a rogue agent that can’t be switched off. But it’s not funny if an agent has been installed at the request of management to solve a corporate problem, and the agent decides the breadth and depth of interaction with the organisation’s systems and data, or as is possible, is not treated with the appropriate security controls to prevent it from interacting with external LLMs and sharing corporate and personal data outside the company – ‘Most agents use a small set of closed-source frontier models’ – through MCP.
Now I understand why the insurance industry is really concerned about the adoption of AI, pricing of AI insurance, and some insurers have introduced exclusion clauses.
Obviously, under the EU Act, models created in the EU have undergone risk assessment, and have closed gaps in governance, risk, and security…………..
@Thaddeus Dziekanowski Brian D. McCarthy Adam Gladsden
@Parava Security Solutions
#artificialintelligence
https://lnkd.in/ehFH7e52
View Original Post
https://www.linkedin.com/feed/update/urn%3Ali%3Ashare%3A7431994857776132096