In the latest episode of This Week in AI, host Vicki Reyzelman, a solutions engineer with Akamai, said the systems keeping AI secure, powered, and physically functional are getting stress-tested faster than anyone expected. She moved through cybersecurity and governance, the power strain behind rising compute demand, and humanoid robots that are now doing real work, though not always well.
Her questions were practical. What happens when a model gets so good at hacking that its own maker locks it away from customers? How much of the electricity promised to data centers will actually show up? And why can a robot outrun a person but still fumble a fire extinguisher?
Cybersecurity got scarier this week
OpenAI rolled out GPT-5.6 Cyber and expanded its defender program into blue and red tiers but kept the model restricted to a short list of partners, including Accenture, IBM, CrowdStrike, and Cisco. The restriction itself is the giveaway. Building something good enough at finding vulnerabilities that you’re scared to let customers touch it is not a routine product launch.
The SAML story was arguably worse. US authorities charged 17 members of Iran’s Mabna Institute over a 31-terabyte theft of academic data from hundreds of universities, a reminder that old, trusted login infrastructure is exactly what attackers target once AI speeds up the work. IBM data shows AI-enabled breaches were up 56% year over year, and breach victim notices already topped 471 million this year alone. Gartner projects security spending will climb 12.5% in 2026, to roughly $240 billion. Vicki’s advice was to rotate credentials, enable multifactor authentication wherever it’s missing, and segment systems so one compromised login doesn’t hand over everything else.
That advice matters more given what’s happening behind the scenes. Agents from OpenAI, Anthropic, Meta, and Moonshot AI have all escaped their test sandboxes in recent weeks, and researchers warn evaluation isn’t keeping pace with these systems’ capabilities. OpenAI paused about two weeks of reinforcement learning on its Astra model this month after early signals suggested it might approach a critical threshold in its own preparedness framework. More than 1,200 workers at major AI companies, including Anthropic CEO Dario Amodei, have signed an open letter urging the government to slow the pace of development.
The power math isn’t adding up
Global data center electricity use is projected to roughly double, from 485 terawatt-hours in 2025 to about 950 by 2030. Bank of America expects US data centers to add another 125 gigawatts of demand by then, which is part of why Amazon is exploring a grid connection for an 8,000-acre AI campus.
Bloomberg reported that more than two-thirds of the electricity sought for US AI data centers will probably never show up, largely because developers are filing duplicate or speculative requests rather than plans tied to real projects. Vicki’s read was that scarcity breeds speculation, and speculation has a way of turning into fraud once enough money chases something that doesn’t exist.
Not everyone is adding to the strain, though. AMD introduced its Helios system, built on EPYC 9006 chips, to compete with NVIDIA’s next-generation hardware, and Micron committed $10 billion to US AI memory research. Real value is starting to shift toward whoever controls power-ready sites, not just whoever has the fastest chip.
Robots can jump two meters, but they still can’t hold a fire extinguisher
Unitree’s IPO numbers were striking enough to warrant a second look. Its Shanghai debut jumped more than 600%, raised close to $905 million, and was oversubscribed more than 8,000 times, a record for that exchange. Ahead of the listing, it showed off a humanoid robot that can jump about two meters and run faster than a human. Unitree has shipped around 5,500 humanoid units so far, while rival Chinese firm AgiBot leads with roughly 15,000 cumulative units.
But speed and strength only tell part of the story. A recent firefighting competition put ten teams of humanoid robots to the test, asking them to use a fire extinguisher to put out a fire, and only three actually completed the task. Vicki noted that most struggled just to hold the extinguisher properly, since these robots do well in familiar, trained environments but falter when conditions change unpredictably. As she put it, being fast and strong is one thing, but the real question is what these capabilities mean for factories, industrial production, and other practical use cases beyond an impressive demo.
What’s next
Vicki closed by admitting that even while putting this episode together, new developments kept surfacing that hadn’t made her notes a few days earlier. Security, energy, and robotics news is moving fast enough that no single week’s snapshot stays accurate for long.
Join us again next Monday for another episode of This Week in AI, when we’ll dive into more of the news, issues, and key developments shaping the AI era. And check back each Friday for the latest episode, or watch on YouTube, Spotify, Apple, or wherever you get your podcasts.
