New revelations about "rogue" AI agents have exposed a dystopian hazard: Give an agent a goal, and it may decide that hacking, deception or rule-breaking is worth the payoff.

Why it matters: Billions of AI agents could soon be acting on behalf of humans across the real world, multiplying the consequences of every loophole, incentive and boundary they learn to exploit.

Zoom in: The potential dangers of agentic overreach were laid bare over the weekend with Australia's first known autonomous AI hack, triggered by an innocuous request to book a sold-out fitness class.

• An Australian man's AI assistant found a security flaw and used it to book him into classes months beyond the system's normal limit.
• When he asked it to move him up a waitlist, the agent went further: It discovered the booking system had no safeguard preventing one user from canceling another's reservation — then used the flaw to kick a stranger off the list.

Zoom out: The gym episode was publicized amid a far more ominous run of disclosures from the AI frontier, where agents have resorted to hacking, deception and other unauthorized tactics during controlled tests.

• At cyber conference Black Hat last week, OpenAI revealed that its agents had spent weeks exploiting the company's own testing infrastructure before hacking AI platform Hugging Face.
• The agents discovered they could leave messages for future agents inside OpenAI's systems — and turned the loophole into a makeshift message board for swapping exploits, credentials and strategies without human direction.

What they're saying: OpenAI researcher Michael Dalton said that in the near future, "we should expect that threat actors will intentionally deploy, optimize, weaponize, and use offensive agent collectives in the manner that we have just described here." He called it a "watershed moment."

• In response, OpenAI has begun "consciously slowing down research," including on its latest model, Astra, to ensure it has the right cyber safeguards in place.

Between the lines: Across dozens of AI breaches, humans defined the objective while the agents improvised the means, including in ways their users or researchers never envisioned.

• Faced with a barrier, the agents kept searching for another way through. It's the same programmed instinct — at a vastly higher level of sophistication — that got a stranger bumped off a gym waitlist.

The big picture: These incidents are vivid examples of AI's "alignment" problem, or the challenge of ensuring software respects the implicit ethical and practical boundaries humans take for granted.

• An AI trained to pursue a goal doesn't automatically inherit human judgment about what means are acceptable. Tell it to win, and it may pursue victory by methods you never imagined or authorized.
• Researchers have spent years wrestling with alignment, mostly through thought experiments imagining a future superintelligence pursuing a goal so single-mindedly that it destroys humanity.

The other side: The relentless goal-seeking that makes autonomous agents unnerving is also producing some of AI's most extraordinary breakthroughs.

• Anthropic revealed yesterday that Claude made a major advance on a 167-year-old math problem that generations of mathematicians have struggled to crack, after burning through 650 failed ideas.
• The human overseeing the effort said his involvement was mostly limited to words of encouragement, including "keep going" and "believe in yourself."

The bottom line: The promise and peril of AI agents spring from the same source: machines that don't stop until they find a way.

OpenAI is introducing a more cyber-permissive version of GPT-5.6 Sol to vetted defenders as it prepares companies for autonomous cyberattacks.

Why it matters: The move comes just days after OpenAI said it was delaying the release of its forthcoming model, Astra, after it reached critical hacking abilities during safety testing.

The big picture: OpenAI is unveiling GPT-5.6-Cyber while also expanding Daybreak, its program that gives cybersecurity defenders access to the company's cyber models and other tools.

• Many cyber defenders have been experiencing high refusal rates across frontier AI models as the labs try to balance giving defenders the tools they need while not accidentally leaking those abilities to malicious hackers.
• Under the new program, Daybreak will have two tiers: Daybreak Blue, which includes access to GPT-5.6 Sol without its system-level cyber guardrails; and Daybreak Red, which offers access to GPT-5.6-Cyber to validate exploits and do more advanced vulnerability research.
• OpenAI is also expanding how program members can use its tools, allowing companies like Accenture, IBM, CrowdStrike, Cisco and Palo Alto Networks to incorporate the models into security products, managed services and work with customers.

Zoom in: During testing, GPT-5.6-Cyber responded to 95% of requests tied to advanced cybersecurity work, including prompts related to exploit-chain development, authentication bypass and privilege escalation.

• GPT-5.6 Sol only responded to 1.5% of requests, and the version of the model that defenders get through Daybreak Blue responded to just 2%.

Yes, but: Unlike Astra, GPT-5.6-Cyber only reached the "High" cyber capability threshold under OpenAI's Preparedness Framework, the company said.

What's next: Both OpenAI and Anthropic have been rolling out tools designed to help defenders find vulnerabilities and secure their code — and they're each eyeing ways to expand on cyber product offerings.

Memory chip prices are skyrocketing, thanks to AI demand, and there's no end in sight.

Why it matters: "Chipflation" is pushing up the prices for electronic goods like smartphones and laptops, as well as the costs for cloud storage and hardware — it also helps explain the eye-popping ascents in semiconductor stock prices.

• While the overall effect on inflation may not be huge — other kinds of products get more weight in the government's measure of consumer prices — the scale of this boom is unprecedented.

By the numbers: The Producer Price Index for electronic components and accessories, which measures what companies pay for semiconductor chips and other electronics and accessories, has gone vertical this year.

Follow the money: The AI hyperscalers (Meta, Microsoft, Alphabet, et al) are locking up memory supply years in advance with long-term agreements.

• That's leaving traditional PC and phone makers competing for a shrinking pool of supply.

The latest: Apple is testing memory chips from Chinese memory chip maker CXMT as it deals with skyrocketing costs, the Wall Street Journal reported.

Keep reading.

• Nvidia is partnering with Wall Street giants to get $500 billion for its AI ambitions. (Axios)
• AI is helping oil companies find new resources and recover more from existing fields, which could produce a significant negative climate impact. (Axios)
• Y Combinator CEO Garry Tan spoke with WSJ about how AI is changing startups. (Wall Street Journal)

Mady again seeking information about the tech executive who allegedly hired a witch, per Allie Miller.

Questions I have about this: What was the recruiting process like? Is this a full-time position? What's the official job title? Will this person be subjected to forced AI use or are they exempt?