Firefox bug breaks privacy resets & LinkedIn gets a new CEO - Tech News (Apr 23, 2026)

First up, a privacy fix you’ll want to know about. Researchers disclosed a Firefox vulnerability where the order of results from an IndexedDB API could act like a stable identifier—basically a high-entropy fingerprint—letting unrelated sites correlate activity without cookies. The worrying twist: in Private Browsing, the signal could persist after closing private windows if the browser process stayed alive, and in Tor Browser it could even survive a “New Identity” reset under certain conditions. Mozilla says it has patched the issue in Firefox 150 and ESR 140.10.0, and other Gecko-based browsers need to ship the mitigation too.

Staying with security, the U.K.’s National Cyber Security Centre is warning that the most serious cyber risk to Britain now comes from hostile states, not just criminals. The NCSC’s Richard Horne called out Russia, Iran, and China, arguing that in a wider conflict the U.K. could be hit “at scale” online—meaning disruption could spread quickly across business systems and critical services. Officials also pointed to a practical problem: AI is helping attackers find weak points faster than many organizations can patch them, making resilience planning less optional and more urgent.

In Microsoft land, LinkedIn is getting a new CEO effective immediately. Dan Shapero, who’s been LinkedIn’s COO for five years and has been at the company since 2008, is taking over from Ryan Roslansky. Roslansky isn’t leaving—he remains a Microsoft executive vice president and is picking up additional responsibilities tied to the Office productivity group. The broader signal here is continuity plus tighter integration, as Microsoft pushes AI features across Office and LinkedIn while it pours money into AI data centers—and wants the org chart to match that strategy.

OpenAI is also pushing the workplace deeper into “agents,” not chats. It’s rolling out Workspace Agents inside ChatGPT for Business, Enterprise, Edu, and Teachers—letting organizations build reusable agents that can operate across tools like Slack, Google Drive, Microsoft apps, Salesforce, Notion, and Atlassian. The key shift is persistence and follow-through: these agents can keep state, run multi-step tasks in the cloud, and keep working even when you log off. OpenAI is clearly betting that the next enterprise battle is governance—controls, approvals, audit trails—because companies won’t deploy fleets of AI helpers without being able to prove what they did and why.

Google, meanwhile, wants your browser to become the workplace control panel for AI. At Cloud Next 2026, it pitched Chrome as an “agentic” platform, including features that can complete multi-step web chores—things like filling forms or scheduling—while using guardrails like confirmations for sensitive actions and boundaries around which sites it can touch. Chrome is also adding reusable one-click workflows and a persistent Gemini side panel across Workspace apps. The business angle is simple: if knowledge work happens in the browser, Google wants AI work to happen there too—under enterprise policy, not in random tabs and unofficial tools.

Now to the AI-coding arms race—because it’s starting to reshape how big tech builds software. Google says about three-quarters of its newly created code is now generated by AI and then reviewed by humans, a steep jump from a year and a half ago. It’s being framed internally as “agentic workflows,” where AI does more of the first-pass implementation and developers focus on steering and validating. The interesting subtext: this changes incentives, team throughput, and even what ‘good engineering’ looks like—less typing, more judgment.

GitHub Next is making a related point: coding agents are still built like “single-player” tools, even though software is a team sport. Researcher Maggie Appleton previewed a prototype called Ace, designed around shared context—think a collaboration space where people and agents plan together, keep decisions attached to the work, and reduce the ‘fast but misaligned’ problem that shows up as painful pull requests and overloaded review queues. Whether Ace becomes a product or not, it captures a growing realization: speed is cheap now; alignment is the bottleneck.

Hiring is shifting too. Startup Sierra says traditional interviews test skills that matter less when candidates can use powerful coding agents. So it replaced classic algorithm-style rounds with an “AI-native” session that mirrors real work: scoping a small product, building it with whatever AI tools the candidate prefers, then defending decisions and showing a path to production. The takeaway isn’t that interviews get easier—it’s that employers are trying to measure judgment, product thinking, and execution under real constraints, because AI can generate code… but it can’t guarantee good decisions.

Let’s talk about the most eyebrow-raising corporate chess move in AI coding: Cursor and SpaceX. Reports say SpaceX has secured the right to acquire Cursor later this year for a huge price—or pay a hefty fee to walk away—while co-developing coding and “knowledge agent” models in the meantime. Separately, Microsoft reportedly explored acquiring Cursor but didn’t make an offer. The big picture: developer tools have become strategic territory. Whoever owns the workflow where code gets written can harvest feedback, iterate faster, and potentially shape the next generation of AI agents.

That brings us to SpaceX itself, where the storyline is getting more complicated ahead of a potential IPO. Elon Musk has been publicly emphasizing projects like orbital AI data centers, moon-based factories, and even AI chip production—while the company’s Mars narrative has taken a back seat. Supporters argue it’s a practical funding ladder; critics see strategy drift. Either way, it matters because IPO investors will be buying not just rockets, but a long-term roadmap—and the clearer that roadmap is, the easier it is to price trust.

On the hardware side, TSMC says it can keep pushing chip advances without jumping early to the newest, pricier generation of lithography machines. Instead, it’s emphasizing a mix of incremental scaling and advanced packaging—combining multiple chiplets and memory into single, high-performance packages. This is one of the industry’s most important themes right now: if classic ‘shrink the transistor’ progress slows, performance gains increasingly come from how you assemble and connect pieces together.

Google is also leaning into custom silicon, unveiling a new TPU generation split into training-focused and inference-focused chips. The company’s pitch is aimed at the “agentic era,” where lots of specialized AI helpers run continuously and cost efficiency matters as much as raw speed. Translation: Google wants to lower the cost of running AI at scale—both for itself and cloud customers—while competing in a world still heavily dominated by Nvidia.

In robotics and transportation, we’ve got three different versions of ‘the future is arriving, slowly.’ Tesla says it now expects 2026 spending to top $25 billion, aimed at scaling Optimus humanoid robots, expanding AI work, and developing its Cybercab autonomous vehicle—an aggressive bet as its core auto business has been under pressure. Amazon struck a deal with Einride to bring more heavy-duty electric trucks into its freight network, pairing vehicles with charging sites as it tries to decarbonize the hardest part of logistics. And in China, Xpeng’s flying-car unit showed off a sizable factory and production-intent builds—one of the clearest signs yet that at least one ‘flying car’ effort is trying to move from spectacle to manufacturing, even if regulation and limited range keep it niche for now.

Finally, two science stories worth your attention. Neuroscientists mapped long-range networks of astrocytes—brain ‘helper’ cells—across the mouse brain, suggesting these cells may connect distant regions in ways that look more circuit-like than we assumed. It’s early, but it hints at an entire layer of brain coordination we’ve been underestimating. And in the U.K., a six-year-old girl’s eyesight was restored through a one-off gene therapy delivered via the NHS—an encouraging reminder that for certain inherited diseases, early diagnosis plus targeted treatment can be genuinely life-changing.