AI bots overtaking web traffic & China’s post-quantum crypto push - Tech News (Mar 20, 2026)

Let’s start with that web-traffic prediction. Cloudflare CEO Matthew Prince says AI bot traffic is on track to surpass human internet traffic by 2027. The key idea is that AI agents don’t browse like we do: one “simple” task—like shopping or research—can trigger requests across thousands of pages and services. For website operators, that’s not theoretical. It’s real load, real bandwidth, and real costs. The bigger takeaway is that the internet may need new norms—possibly even new infrastructure—so automated visitors can do work without overwhelming or exploiting the sites they depend on.

Now to security and geopolitics, where the next encryption shift is getting closer. A leading Chinese cryptographer, Wang Xiaoyun of Tsinghua University, says China is likely to develop national post-quantum cryptography standards within the next three years. The driver is straightforward: powerful future quantum computers could break much of today’s widely used encryption, which would force governments and major industries to migrate to new algorithms. Wang also suggested China may favor a different mathematical approach than the one underpinning international standards—an early signal that the “standards race” isn’t just academic. If major economies diverge, global cybersecurity could become more fragmented, with real consequences for hardware, software, and cross-border trust.

That standards competition is already visible elsewhere. The U.S. finalized its first post-quantum cryptography standards in 2024, and it’s aiming for broad migration by the mid-2030s. What’s interesting here is the timeline mismatch: migrating cryptography is slow, and some data needs to stay secure for decades. That’s why finance and energy keep coming up as early priorities—they’re high-value targets with long-lived sensitive information.

Developer tooling had one of the biggest business headlines: OpenAI says it will acquire Astral, the company behind widely used Python tools including uv and ruff, with the team joining OpenAI’s Codex organization. The immediate promise is continued open-source support, but the broader story is about control of infrastructure. Tools that manage dependencies and workflows quietly become foundational, and ecosystems get nervous when “load-bearing” components consolidate under a single corporate roof. The upside is faster development and funding stability. The risk is subtle: governance, priorities, and competitive incentives can shift—and the community will be watching closely for signs of leverage or lock-in.

Staying with software, two widely shared pieces this week tried to bring sanity to AI-assisted coding. One, a manifesto-style guide called “AI Code,” argues that maintainability is the new battleground when multiple AI agents can generate piles of code quickly. The message is to be intentional: keep core building blocks small and reusable, keep messy workflow code clearly separated, and model data so invalid states are hard to represent. The other, from engineer Edward Z. Yang, focuses on the human side—especially junior developers—who can feel buried by the need to understand every line an AI spits out. His advice is: read less, steer more. Treat the model like a fast typist that needs strong direction, not a teammate you blindly trust. The common thread is oversight: speed only helps if teams keep clarity and control.

Meanwhile, the bigger AI debate keeps escalating. One long essay argues AI already beats humans across most cognitive domains, and that the limiting factor is no longer raw capability—but organization: tooling, workflows, and “scaffolding” that lets systems execute multi-step work reliably. Another critique points out how muddled the conversation has become, with the term “agent” often used as marketing more than a description of what a tool actually does. The practical takeaway for businesses is familiar: real value tends to show up when processes are redesigned around new capabilities, not when companies try to swap humans for software one role at a time.

On the research frontier, momentum is building around what’s being called “world models.” These systems aim to predict how an environment changes when an agent takes an action—less about generating pretty outputs, more about building an internal sense of cause and effect. Why it’s interesting: if AI can learn reliable action-and-reaction patterns, it can “practice” decisions in a learned environment before doing anything risky in the real world. That could accelerate robotics, driving, and other embodied AI—though data is a bottleneck, because truly useful training requires sequences paired with the actions that caused them.

Now to supply chains, where a conflict far from chip fabs can still ripple straight into them. Analysts are warning that disruptions in Qatar—previously a major helium supplier—could tighten global helium availability after attacks affected operations at Ras Laffan, a key industrial hub. Helium isn’t just party balloons; it’s a critical input for semiconductor manufacturing and for medical imaging. If the disruption drags on, chipmakers—especially those relying heavily on Gulf imports—could see intermittent shortages and higher costs. Even if big customers get priority, the broader lesson is that advanced manufacturing still depends on a few fragile upstream links.

On the policy front, Europe is moving to clamp down on one of the ugliest corners of synthetic media. Key European Parliament lawmakers have backed adding a ban on AI “nudification” apps—tools used to create explicit images without consent—under the EU’s AI Act. At the same time, lawmakers also appear open to pushing back some compliance deadlines for high-risk AI rules, largely because the technical standards may not be ready. The direction is clear: stronger protections against non-consensual sexual deepfakes, paired with a reality check on how fast complex compliance frameworks can be implemented.

Brazil also rolled out a major new digital rule set for minors. Its Digital Statute of Children and Adolescents has taken effect, tightening how social platforms and digital services treat young users. The law pushes for stronger age checks, more guardian involvement for younger teens, and it targets engagement mechanics that keep kids scrolling. The larger significance is that child safety is shifting from “parental controls” toward direct platform responsibility—something more countries are likely to test in the coming years.

Turning to space, a defense startup called True Anomaly is making a bet that future orbital conflict looks more like slow, deliberate close-approach maneuvers than cinematic dogfights. The company is building maneuverable satellites designed to approach, inspect, and shadow objects in orbit—capabilities the U.S. Space Force is actively exercising, including scenarios where one spacecraft plays the adversary. The interesting part isn’t just the hardware; it’s the doctrine shift. Space is being treated less like a pristine environment of expensive one-off satellites, and more like an operational domain where scalability, responsiveness, and repeatable missions matter.

In biotech, researchers reported progress toward a long-held goal: generating CAR T cells directly inside the body, rather than manufacturing them patient by patient. In early models, the approach achieved targeted insertion at a specific T-cell locus, producing meaningful therapeutic effects without some of the systemic reactions that worry clinicians. This is still early-stage science, but the significance is big: if therapies that currently require complex, bespoke manufacturing could be delivered more like a drug, it could change cost, access, and speed—three of the biggest barriers in cell therapy today.

And for a very different kind of “living tech,” researchers in Singapore built lab-grown skeletal muscle that effectively trains itself by contracting against a paired muscle—like a built-in gym. They used it to power a biohybrid swimming robot that set a speed record for this category and could be controlled more precisely than many earlier muscle-driven systems. It’s a niche corner of robotics, but it addresses a real bottleneck: cultured muscle is often too weak for practical machines. Stronger biological actuators could eventually enable soft, efficient, and even biodegradable robots for environmental monitoring or short-term medical applications.

Finally, a quick look at two China stories that point in the same direction: accelerated adoption of frontier tech, paired with tighter control. First, brain-computer interfaces. China is stepping up invasive BCI commercialization and trials, and state-backed players are openly benchmarking themselves against Neuralink. The near-term focus is medical—restoring motor function and enabling assistive control—but the strategic investment signals this is becoming a national priority area. Second, on AI agents: OpenClaw, an open-source computer-using assistant, is spreading rapidly in China with help from public events and local incentives. Cities are even courting “one-person companies”—solo founders using AI tools to run operations that used to require a team. But alongside the enthusiasm, authorities are also warning sensitive sectors to limit usage over data and security risks. In other words: scale fast, then regulate hard.