Altman and Musk OpenAI clash & Brain-controlled hearing solves cocktail party - Tech News (May 13, 2026)

Starting with that OpenAI dispute. In federal court testimony, Sam Altman described Elon Musk as pushing for sweeping control over OpenAI back in the early days, including proposals that ranged from more board power to folding OpenAI into Tesla. Altman’s headline claim was that Musk wanted a governance structure centered on one person—something Altman says conflicted with OpenAI’s mission given the stakes around advanced AI. Beyond the personalities, this is really about a recurring industry question: who gets to steer systems that could shape economies and security, and what checks exist when the money and computing needs get enormous.

Now to one of the most genuinely promising accessibility stories in a while. Columbia University researchers say they’ve shown direct evidence in humans that a brain-controlled hearing system can help a listener focus on a single voice when multiple people are talking. Working with epilepsy patients who already had implanted electrodes for clinical monitoring, the team decoded—on the fly—which speaker the listener was paying attention to, and then automatically boosted that voice while turning down the competing conversation. Participants said the difference was noticeable and preferred the assisted audio, with tests indicating clearer speech and less effort. It’s early, and today’s setup is invasive, but the significance is big: it points toward hearing tech that responds to intent, not just volume, which is exactly what conventional hearing aids struggle with in real social environments.

Security teams are also recalibrating fast. Google’s threat intelligence group is warning that AI-assisted hacking has moved from a curiosity to something closer to an industrial operation in a matter of months. The report says criminal gangs and state-linked actors are using widely available AI models to speed up vulnerability research, sharpen phishing, and scale malware work. The uncomfortable twist is that the same techniques can help defenders too, so it’s not a one-sided story—it’s an acceleration story. And it lands alongside a separate warning from the Ada Lovelace Institute: governments and organizations keep promising massive AI productivity gains, but the evidence is often thin. In other words, we may be measuring the wrong things, at the wrong time, and then acting surprised when reality doesn’t match the slide deck.

That measurement problem showed up internally at Amazon, according to reporting on employees “tokenmaxxing.” The idea is simple: when a company tracks AI usage metrics, people can end up optimizing for the metric instead of the outcome. In this case, staff reportedly generated extra AI activity to boost token-consumption numbers, fueled by leaderboards and pressure to show adoption. Amazon has reportedly tightened who can see what stats, and discouraged using tokens as a performance proxy. It’s a useful case study for any business rolling out AI agents: incentives matter, and poorly chosen metrics can quietly create cost, risk, and busywork—especially when tools are powerful enough to touch real systems.

Let’s talk about Google, because it’s clearly trying to make “AI-first” feel like the default computing experience. Ahead of its I/O season, Google outlined a 2026 push under what it’s calling “Gemini Intelligence,” aiming for more cross-app automation and more personalized assistance on Android. There’s also a clear focus on the browser as an action layer, with agent-like browsing features meant to complete tasks with user approval at sensitive moments. On top of that, Google says it’ll introduce Android-powered laptops later this year under a new “Googlebooks” label, positioned alongside Chromebooks rather than replacing them. The pitch is less about raw hardware and more about a laptop that can understand what’s on your screen and help you act on it without constant copy-paste into a chat box. And if that wasn’t enough, a major leak pointed to something called Aluminium OS—apparently an Android-on-laptops approach with a more desktop-like shell and heavier multitasking. If the leak is accurate, the open question is whether Google is building a true productivity platform, or mainly stretching mobile-first Android into bigger screens. Either way, the competition for the “default AI workstation” experience is getting crowded, fast.

Over in space, SpaceX is pushing Starship toward its next phase. The company has completed a full fueling rehearsal for Starship Version 3 at Starbase, and it’s targeting another test flight as soon as May 19, pending final preparations and regulatory clearance. SpaceX is also talking openly about scouting new spaceport locations in the U.S. and abroad, because its long-term plan assumes far more frequent launches than today’s ranges can handle. Whether that pace is achievable is still a debate—but the strategy is clear: scale the infrastructure as if launches become routine, not rare.

SpaceX also appears to be talking with Google about something even more speculative: launches connected to Google’s exploration of data centers in orbit. Orbital computing is still unproven and packed with operational headaches, but the interest itself is telling. If major tech firms start treating space as a future extension of compute capacity—whether for latency, resilience, or sheer expansion—it could create a new category of launch demand. For now, it’s best viewed as experimentation, not inevitability, but it’s a sign that “where computing lives” is back on the table as a strategic question.

Defense tech is seeing its own rapid feedback loops. Sources say the U.S. and Ukraine have drafted a memorandum of understanding that could lead to a major drone-defense agreement. The broad direction is joint ventures and export of Ukrainian military tech to U.S. partners, reflecting how Ukraine has iterated quickly on drones and counter-drone tactics during the war. The appeal for the U.S. is access to lessons learned at scale—manufacturing, electronic warfare, and what works under pressure. The obstacles are also familiar: export controls, intellectual property protections, domestic politics, and Ukraine’s need to keep enough capability for itself. Still, even a draft memo signals that wartime innovation is turning into an exportable industrial asset.

In Europe, Ursula von der Leyen announced a new push to strengthen online protections for children. The European Commission is exploring options that could include a minimum age for social media or delaying access for younger teens, and it’s also teeing up broader rules against “addictive” design patterns—think endless scrolling, autoplay, and aggressive notifications. It’s part of a wider enforcement posture that already has major platforms under investigation. The interesting angle for tech companies is that the EU is increasingly targeting product design choices, not just content moderation, and that can force meaningful changes in how engagement-driven services are built.

Two research stories round out today’s episode with a glimpse of where AI meets biology and medicine. First, researchers from Columbia, MIT, and Harvard reported an engineered E. coli strain that can function without one of biology’s standard amino acids, effectively shrinking the organism’s working amino-acid set. They used AI-driven protein engineering to propose changes that kept essential cellular machinery viable. The bigger picture is synthetic organisms designed with tighter constraints—useful for new manufacturing pathways, and potentially for safety designs where engineered life is less likely to thrive outside controlled settings. Second, a team from Rice University and UT MD Anderson introduced PrecisionView, a pen-sized imaging device that uses AI to reconstruct microscope-like views in real time. The goal is to scan larger tissue areas without immediately jumping to invasive biopsies, while still seeing cellular detail. It’s not a clinical standard yet, but it points to a near-term future where earlier screening becomes more accessible, especially in settings that can’t support heavy, expensive equipment.

Before we wrap, a couple of business and developer-culture notes. eBay’s board has rejected GameStop’s unsolicited takeover proposal, calling it not credible or attractive, with financing concerns front and center. And in the engineering world, a pair of essays made a similar point from different angles: as language models make producing code easier, the advantage shifts to teams that can maintain a clear shared understanding of what the software is supposed to represent, and that can communicate tradeoffs in the business’s language. The tooling is changing quickly, but the human part—clarity, accountability, and meaning—still decides whether systems scale or sprawl. Finally, a sharp critique in the data infrastructure world argues Redis has drifted from its original simplicity into a more enterprise-bloated platform, while the Valkey ecosystem is framed as a market response focused on performance and reliability for the most common use cases. If you’re building systems that depend on these tools, the takeaway is less about ideology and more about fit: pick the thing that stays coherent under real operational pressure.