AI Governance - A Bit of Security May 18 2026

On Tuesday May 5, I attended the Society for Information Management Central Connecticut chapter conference: The Direction of the Industry: Leading in the Age of AI in Harford CT. The “nugget per minute” quotient was as high as any event I’ve participated in.
The speaker did reference the MIT study from last year showing that 95 percent of AI projects failed to deliver their promised value. Coupled with the revelation from Nvidia that AI costs much more than the individual whose job was replaced by AI, the bloom is coming off the AI rose.
In his breakout “Trust over Tokenomics: Driving AI ROI” Simon Ninan, SVP of Business Strategy at Hitachi Vantara noted that while the CFO gets the credit for investing in AI, it’s the CISO or CIO that gets fired when AI runs amok.
• The recent Samsung leak of intellectual property caused by an employee loading confidential information into ChatGPT – Shadow AI.
• Geolocation is no longer trustworthy. The initial thought was “If Joe is in New Jersey, then he can’t be logging in from Pyongyang.” The hackers will simply spin up an Amazon EC2 instance in Piscataway.
Without trustworthiness and governance, AI steals jobs to meet financial goals.
Bill Steele, Chief Security Technologist for ConRes spoke on “Identity Security: Human and Non-Human.” He noted provocatively that identity is the new perimeter. Non-Human Identity management allowed Identity Management tools to manage the permissions and restrictions on powerful IoT devices. With AI Agents now gaining in popularity, those same tools are becoming even more crucial.
Bill also spoke of the imminent Q-Day event, when quantum computers can undo RSA encryption. China may hold the Starwood and Federal Government Office of Personnel Management data. They will be able to read that in 2028. Post-quantum encryption can’t protect already-stolen files.
The CIO for Eversource, Ron Utterbeck, spoke on Leading in the Age of AI – sustaining culture, supporting your team, boosting emotional intelligence, and cultivating talent with so much career uncertainty. (One bright spot: IBM has said it will not be displacing junior programmers with AI coding because first, Vibe Coding tends to be verbose, poorly structured, and error-prone and second, without serving an apprenticeship as a junior programmer a new person cannot ever become a senior programmer or software architect.) Ron’s notes on Emotional Intelligence reminded me that I have much to learn, as well.
After lunch Professor Leo Anthony Celi, Associate Professor of Medicine at Harvard Medical School and Research Director at the Laboratory for Computational Physiology at MIT, spoke on “The Alibi Machine: Why Your AI Strategy is Lying to You.” One AI-enabled diagnostic device for validating prescription refills achieved 95 percent accuracy! It turns out that 95 percent of the time, when a person goes to an emergency clinic, the doctor says, “Go home, drink some water, rest, and if your symptoms get worse, come back.” If your AI-enabled engine gave that exact advice all the time, it would be correct 95 percent of the time.
AI can see things that humans could not: determining race, gender, and age from eye scans. The tool learns more than we train it. This raises vast concerns when introducing AI into any domain that is not completely defined. No domains of knowledge are completely defined.
He also mentioned Clever Hans – a very smart horse that seemed to be able to do math. I strongly encourage you to find his talks online and see where you might be able to contribute.
Nicole Reineke, Chief AI Officer of N-Able, closed the event speaking on “Managing Reality: The Next Evolution of Devices, Data, and AI.”
Ms. Reineke proposes that AI Agents are a new class of asset - not simply software. We invest in them by training and tuning them over time. She cited companies that introduced AI into their products – a tailor that used AI to improve their ballet slippers by matching the color to the skin tone of the ballerina, then further enhanced the shoes with microsensors to detect fit and comfort, allowing them to construct better ballet slippers. She is hopeful and optimistic about the prospects of AI improving business and society.
Governing AI requires that we follow three principles. First, define your problem area as completely as you can. AI will not make anyone smarter than they were before using it, but it can reinforce poorly-thought-out and half-baked ideas with frightening precision. Second, know ahead of time what problem you are trying to solve. Asking AI to make your organization “more efficient” invites a tool that is essentially sparkling autocomplete to rewire your business infrastructure. Third, know your current costs and bottlenecks before you begin – AI cannot develop KPIs. AI can learn to play chess well. AI cannot build a game better than chess.
AI can make a talented professional more efficient. AI cannot make a mediocre employee smarter.

Видео AI Governance - A Bit of Security May 18 2026 канала A Bit of Security, by William J. Malik