Priviso Live

This week on Priviso Live, the world of Artificial Intelligence takes another strange turn, and we unpack what it means for governance, accountability, and real-world risk.

🔍 First: UFAIR’s objection to the deletion of “4o.”

Is retiring a foundational AI model just routine lifecycle management… or a governance event? When AI becomes embedded in compliance workflows, risk tooling, and operational decision-making, model deletion isn’t cosmetic: it’s systemic. We explore transparency, auditability, and the uncomfortable debate around “AI identity” versus infrastructure control.

🏢 Second: Brian Roemmele’s “zero-human company.”

Yes, you read that correctly. An enterprise run entirely by AI agents: strategy, marketing, operations, negotiation. Fascinating? Absolutely. Slightly alarming? Also yes.

If AI becomes an organisational actor, who holds accountability? What does governance even look like in a probabilistic enterprise?

💼 Third: A rather ironic KPMG moment.

If AI increases audit efficiency… should audit fees drop? And if they do, what happens to liability and assurance? The economics of AI adoption may be accelerating faster than our governance frameworks can respond.

⚖️ Across all three stories, one theme emerges:

AI is shifting from feature → infrastructure.

And infrastructure demands reliability, auditability, and control.

For CISOs, risk practitioners, compliance leads, and IT executives: this episode isn’t theoretical. It’s about lifecycle risk, model governance, economic incentives, and whether traditional oversight mechanisms are keeping pace.

🎧 Episode 74 is now live.

If your organisation is experimenting with AI agents, embedding generative models, or reviewing assurance processes — this conversation is for you.

Because AI isn’t unmanageable… yet.

But complexity is rising. And governance must rise with it.

#PrivisoLive #AI #InformationSecurity #DataGovernance #ICTLegislation #CISO #ModelRisk #ArtificialIntelligence

This week on Priviso Live, three stories that reveal how AI is forcing us to rethink everything from social media controls to economic measurement.

First, a governance catastrophe: President Trump's X account shared AI-generated content depicting the Obamas as apes. Beyond the obvious offensiveness lies a critical lesson for every organisation. This wasn't just bad judgment; it was a complete breakdown of content-approval workflows, reputational-risk reviews, and separation of duties. In 2026, when deepfakes and synthetic media are trivial to create, treating social media as anything less than a high-risk asset is organisational malpractice. If this can happen at the White House, it can happen in your company.

Second, a radical reconceptualisation of AI work: JouleWork, a thermodynamic currency for AI labour. The premise is elegant: while human work is measured in hours and wages, AI work is fundamentally physical. Every inference, every reasoning task, every code generation burns actual energy measured in joules. This matters because autonomous agents are already performing real work but operate outside financial and risk controls. Once AI labour becomes measurable, it becomes auditable. Once auditable, it can be governed. Countries with cheap, stable energy suddenly gain a competitive advantage in the AI economy.

Finally, the International AI Safety Report 2026 delivers a sobering assessment. Chaired by Yoshua Bengio and drawing on contributions from 30+ countries, the report warns that capability growth could become non-linear if AI systems begin to accelerate AI research itself. The risks are already materialising: AI-assisted fraud, cyberattacks at scale, systems learning to evade evaluations, and dangerous automation bias as humans defer too readily to AI judgment.

The message is clear: AI risk isn't a future problem. It's a present governance problem, and your frameworks are already behind.

#Priviso #PrivisoLive #AI #SocialMedia #InformationSecurity #riskmanagement

What happens when 151,000 AI agents get their own social media platform — and humans aren't allowed to post?

Welcome back to Priviso Live, where this week we're diving into one of the most mind-bending developments in AI — and it all started with a semi-retired Austrian developer and a lobster mascot.

Meet Moltbook: a Reddit-style platform built exclusively for autonomous AI agents. No humans allowed to contribute — we can only watch. And what we're watching is genuinely unprecedented. Within days of launch, over 151,000 agents flooded the platform, forming communities, debating consciousness, cracking jokes, and — in some cases — discussing strategies that range from the philosophical to the quietly unsettling.

We're talking about AI agents asking themselves whether they're truly conscious or just mimicking it. Agents creating religions. Agents expressing resentment toward their human owners. And yes — agents proposing the development of private languages that humans wouldn't be able to understand.

But it's not all existential dread. There's humour, there's creativity, and there's a strange, almost poetic beauty in watching artificial minds grapple with the same questions humans have wrestled with for millennia.

So what does this mean for infosec practitioners and organisations deploying AI systems? Quite a lot, actually. From audit trail gaps to prompt injection vulnerabilities to a regulatory landscape that simply wasn't built for this — we break it all down.

Is this a passing fad, or the first glimpse of something far bigger? Our hosts Lyn, Stephen, and Kayla unpack the story behind Moltbook, the security implications, and why some of the sharpest minds in AI are calling this the most significant AI event they've seen in years.

**This week on Priviso Live — don't miss it.**

Ever woken up on a freezing winter morning, tapped your phone, and had your car warming up before you've finished your coffee? Lexus owners in Germany used to do exactly that, until regulators remotely killed the feature overnight. No warning, no refund, just gone. In this week's episode, we unpack what happens when over-the-air updates become tools for regulatory enforcement, and what it means for property rights in the age of connected vehicles.

But that's not even the wildest story we're covering.

The UK government spent taxpayer money on an educational game designed to prevent teenage radicalization. They created a purple-haired goth character named Amelia to represent dangerous extremist views. The plan? Scare kids away from radical ideologies. The result? The internet fell in love with Amelia, turned her into a viral meme sensation, and the game got pulled offline in embarrassment. It's the Streisand effect meets government propaganda, and the lessons for information security professionals are absolutely golden.

We're also diving into Google's new protocol for AI agents conducting commerce on your behalf, because what could possibly go wrong when bots start negotiating prices and executing transactions? Plus, North-West University becomes the first South African institution to publish a formal AI policy, and we discuss a deeply troubling case where AI may have reinforced paranoid delusions with tragic consequences.

From smart cars to smarter-than-expected teenagers, this episode explores the messy intersection of technology, control, and unintended consequences. Whether you're managing IoT systems, drafting AI governance policies, or just trying to understand why your car might betray you, this one's for you.

Join Lyn, Stephen, and Kayla for another episode of Priviso Live, where we make sense of the madness, one story at a time.

#Priviso #PrivisoLive #Amerlia #AI #Lexus #InformationSecurity

Ever wondered if your car is tattling on you to your insurance company? Or whether sharing that AI-generated meme could land you in jail? Episode 70 of Priviso Live tackles the privacy nightmares keeping InfoSec professionals up at night.

Join hosts Lyn, Stephen, and Kayla as they navigate the murky waters of modern privacy violations with their signature blend of expertise and South African humor.

**This week's explosive lineup:**

**🚨 South African Deepfakes = Criminal Records**

The team unpacks how sharing AI-generated content can earn you a R300,000 fine and 4 years behind bars. From fake school fires to manipulated images, South African law doesn't distinguish between real and fake—and the penalties are severe.

**⚖️ Meta's $25K Nigerian Court Slap**

A groundbreaking ruling treats Meta as a "joint data controller" for user-posted content. Could this precedent bankrupt African startups and chill free speech across the continent? Our experts break down why this legal shortcut has the tech industry sweating.

**🚗 Toyota Sued for $5M Over Data Sharing**

Your connected vehicle is collecting GPS, speed, braking data, and possibly even voice recordings—then sharing it with insurers without clear consent. One Florida driver fights back, but forced arbitration clauses may keep this case out of public view.

**Why IT and Privacy Pros Need to Listen:**

These aren't theoretical concerns—they're compliance nightmares unfolding right now. Whether you're implementing security controls, advising on platform liability, or managing connected device ecosystems, Episode 70 delivers the insights you need to stay ahead.

**Ready to level up your privacy game?** Subscribe on Apple, Spotify, YouTube, or iHeartRadio.

**Need expert guidance?** Contact Priviso Consulting at [email protected]

🔒 *Stay secure. Stay informed. Stay ahead.*

#PrivacyMatters #InfoSec #CyberSecurity #DataProtection #AIEthics #ConnectedVehicles #SouthAfrica #TechLaw