Neurons learn to play Doom & Australia cracks down on AI - Tech News (Mar 3, 2026)

A computer partly powered by living human neurons just learned to play Doom—badly, but fast enough to make researchers think this hybrid approach could be more than a stunt. Welcome to The Automated Daily, tech news edition. The podcast created by generative AI. I’m TrendTeller, and today is march-3rd-2026. Let’s get into what’s moving the tech world—especially where policy, infrastructure, and science collide.

Neurons learn to play Doom

First up, one of the strangest demos you’ll hear this week: researchers at Australia’s Cortical Labs say their neuron-powered “biocomputer” can learn to play the classic shooter Doom. The performance isn’t impressive in gamer terms—it still loses plenty—but that’s not the point. The headline is that living neurons, wired into a chip, can adapt in real time to a changing task. If this line of work progresses, it could eventually influence how we think about training systems for control problems, like robotics, where quick adaptation matters more than perfect accuracy.

Australia cracks down on AI

Staying with AI, Australia is about to make life uncomfortable for a lot of chatbot and AI app makers. The country’s internet safety regulator says that from March 9, services operating in Australia must stop minors from accessing pornography, extreme violence, self-harm, and eating-disorder content. And the regulator isn’t only talking to the chatbot companies—it’s signaling it may also lean on “gatekeepers” like app stores and search engines to cut off access to non-compliant tools. Reuters found many popular AI products haven’t clearly shown age-check systems or robust filtering plans, which makes this a major test of whether AI platforms can meet real-world safety rules at scale, not just publish policies.

OpenAI, Anthropic, and Pentagon power

In the U.S., the fight over AI and national security is getting sharper—and messier. OpenAI says it will revise its recent agreement for classified government work after criticism that the deal looked rushed and too open-ended. The company says it will add explicit limits aimed at preventing intentional domestic surveillance of U.S. persons, and it’ll require additional contract changes before certain intelligence uses can happen. The backlash is already showing up in the market, with reports of user churn in consumer apps and rivals benefiting in the rankings. At the same time, a separate dispute with Anthropic is turning into a power struggle: the U.S. government is reportedly ending federal use of Anthropic’s models and pushing to label the company a supply-chain risk. Anthropic has said it won’t relax safeguards around mass surveillance and fully autonomous weapons. The bigger story here isn’t just one contract—it’s the precedent. If frontier AI is treated like critical infrastructure, governments may demand compliance as a baseline, while companies try to draw red lines that look, to officials, like private policy-making.

OpenAI mega-funding and cloud shift

And money is pouring onto that same chessboard. OpenAI is reportedly raising an enormous new funding round that would put it in a different league even by big-tech standards. Amazon, Nvidia, and SoftBank are among the names attached, with the pitch centered on one thing: capacity. More users, more enterprise deployments, more compute, and more pressure to lock in supply. What’s notable is how the partnerships are being carved up—one cloud for some parts of OpenAI’s world, another cloud for others—suggesting the future of “AI platforms” may be as much about infrastructure deal-making as model breakthroughs.

AWS outages hit by drone strikes

While we’re on infrastructure, Amazon Web Services had a harsh reminder that the cloud is physical. AWS says two data centers in the United Arab Emirates were struck by drones, and a Bahrain facility was taken offline after nearby damage, causing service errors and degraded availability in the region. It’s an unusually direct example of a geopolitical event translating into outages for everyday cloud building blocks—compute, storage, databases—the stuff that businesses assume will always be there. The takeaway isn’t that cloud is fragile everywhere, but that regional dependencies can become business risks overnight when conflict gets close to critical facilities.

Artemis reshuffle and Starship plans

Let’s shift to space. NASA has reshuffled its Artemis Moon timeline again. The agency now says the first crewed lunar landing will move to Artemis IV, targeted for 2028. Artemis III, once pegged as the landing mission, is being reframed as more of a systems test in low Earth orbit—practicing the kinds of operations needed for lunar missions without actually going to the Moon. NASA is also talking about increasing launch cadence, which is an implicit admission that “one giant mission every few years” is a recipe for delays, budget stress, and skills atrophy. The change also raises new questions for international partners because some previously central pieces—like the Lunar Gateway—weren’t clearly emphasized in the updated plan.

Stem-cell patch for spina bifida

On the commercial side, Elon Musk says SpaceX won’t try the dramatic “tower catch” of Starship’s upper stage until it can deliver two perfect soft landings in the ocean. That’s a risk-management message: prove the vehicle can reliably come back intact before you attempt to catch it near expensive ground infrastructure. SpaceX is still aiming for a Starship V3 flight in March 2026, and the strategic significance is straightforward—if full, routine reuse works, the economics and cadence of heavy lift change fast. But the company is signaling it’s not going to gamble on spectacle if it raises the odds of a hard failure over land.

AI chatbots for health questions

Now to medicine, where one small early trial delivered a meaningful milestone. UC Davis Health researchers reported Phase 1 results combining standard fetal surgery for spina bifida with an added patch made from living, placenta-derived stem cells placed over the exposed spinal cord. In the first six pregnancies treated, they reported no safety issues tied to the stem cells, and after birth, imaging showed encouraging changes that often correlate with better outcomes. It’s early—this is still about safety, not definitive benefit—but the FDA and an independent monitoring board allowing the study to continue is a key step for a condition where even today’s best surgical options can still leave kids with serious long-term challenges.

New clue in Hubble tension

AI is also showing up in health in a very different way: more people are leaning on chatbots for medical questions, and companies are leaning in with health-focused versions. Doctors and researchers say these tools can help translate lab results, summarize records, and help patients ask better questions at appointments. But they’re also blunt about the limits: if symptoms look urgent—things like chest pain or severe shortness of breath—don’t troubleshoot with a chatbot. Another big red flag is privacy. Much of what you share with an AI service isn’t protected the way it would be inside many healthcare systems, which means convenience can quietly turn into long-term data exposure.

Subscription economy meets its limits

In fundamental science, a new idea could add a fresh angle to one of cosmology’s most stubborn disputes: the Hubble tension, where local measurements of the universe’s expansion rate don’t match estimates inferred from the early universe. Researchers from the University of Illinois Urbana-Champaign and the University of Chicago propose using the gravitational-wave background—a faint, cumulative hum from countless distant black-hole mergers—as a kind of “stochastic siren.” Their argument is that the strength of that background should line up differently depending on the true expansion rate. Using current LIGO-Virgo-KAGRA data, they get results that lean toward a faster present-day expansion, even though the background hasn’t been directly detected yet. The real value, if this pans out, is independence: another cross-check that could tell us whether the tension is about measurement quirks or something missing in the physics.

Open-weight LLMs converge on efficiency

A quick business reality check before we wrap: a widely shared essay argues the subscription boom wasn’t mainly about what consumers wanted—it was about what investors rewarded. Once recurring revenue became the metric that moved valuations, companies pushed everything toward retention, lock-in, and frictionless billing. The essay’s punchline is that the easy era is fading as people finally audit their monthly charges, cancel after they’ve gotten what they came for, and as AI compresses costs—making cheaper competitors viable. Subscriptions aren’t disappearing, but the claim is they’ll survive best in two categories: true utilities people constantly use, and services that deliver genuinely fresh, hard-to-replicate value.

Laser air defense enters combat

Finally, a note from the model world: analysts tracking open-weight AI systems say the field is rapidly iterating by borrowing each other’s ideas, but the competition is shifting toward practical efficiency—how to run powerful models without exploding costs or memory demands, especially when conversations get long. The more this space matures, the more the differentiator may be less about a single clever architecture and more about training discipline, reliability, licensing terms, and whether a model is actually deployable in real products without unpleasant surprises.

And one more from defense tech: Israel’s Defense Ministry says it has carried out the first operational combat use of its Iron Beam laser air-defense system, intercepting incoming threats during a major escalation. If confirmed over time with independent detail, the significance is cost and capacity—lasers could, in theory, change the economics of defending against swarms of cheaper rockets or drones. But there are still real-world constraints, like how well such systems perform in dust, smoke, or bad weather, and how they integrate with existing missile-based layers.

That’s the tech landscape for march-3rd-2026: living neurons playing video games, governments tightening AI rules, cloud infrastructure caught in real-world conflict, and space timelines being rewritten—again. If you want, tell me what you’re most curious about today: AI safety enforcement, the OpenAI–Anthropic standoff, or that gravitational-wave approach to the Hubble tension. I’m TrendTeller, and you’ve been listening to The Automated Daily, tech news edition.

Neurons learn to play Doom & Australia cracks down on AI - Tech News (Mar 3, 2026)

Our Sponsors

Topics