Opinion: The Future World Will Be Like China in 2026

I do not make a moral judgment here about whether mass surveillance is good or bad. This is a prediction, not an endorsement.

My prediction is that the future world will increasingly resemble China in one specific respect: governments will build increasingly comprehensive systems for observing physical and digital activity, identifying potentially dangerous behavior, and intervening before an attack occurs.

By “like China,” I do not mean that every country will adopt China’s political system. I mean something narrower: dense networks of cameras and sensors, biometric identification, automated data analysis, extensive digital monitoring, and security systems designed increasingly around prevention rather than investigation after the event.

The reason, I argue, is technological asymmetry: the destructive capability available to an individual or a very small group is increasing much faster than society’s tolerance for the resulting risk.

When One Person Can Cause Mass Harm

Historically, causing destruction on the scale of a state usually required an army, an industrial organization, or at least a substantial terrorist network.

Technology gradually changes that equation.

Cyberattacks can already be conducted by relatively small teams. Biotechnology creates legitimate scientific benefits while also generating dual-use risks. The World Health Organization explicitly treats the deliberate release of dangerous pathogens as a global health-security concern and notes that deliberate events can potentially produce mass casualties and international public-health emergencies. Its biosecurity guidance also specifically considers emerging technologies, including artificial intelligence.

The trend therefore matters even if catastrophic misuse remains rare:

When the potential damage caused by one actor rises by orders of magnitude, security institutions have incentives to detect dangerous activity before the final act rather than merely identify the perpetrator afterward.

Catching somebody after a deliberately released pathogen has spread internationally may be close to irrelevant as a security strategy. Prevention becomes much more valuable than punishment.

That is the fundamental pressure pushing societies toward surveillance.

“Making People Better” Cannot Eliminate the Risk

One possible response is cultural: educate people better, reduce extremism, improve mental health, decrease inequality, prevent radicalization, and create a society in which fewer people want to cause harm.

All of these measures can reduce risk. I do not think they can reduce it to zero.

Dangerous actions do not arise exclusively from people who possess a stable malicious ideology. They can result from coercion, severe error, intoxication, neurological impairment, temporary loss of judgment, or other abnormal circumstances.

I have a personal reason for taking this possibility seriously. After once being hit in the head, I subsequently behaved irrationally and set a forest on fire. I am not describing that event as terrorism: terrorism normally implies deliberate violence or intimidation in pursuit of political, ideological, or similar goals. My point is different.

A person does not necessarily need to be a committed terrorist to become temporarily dangerous.

Consequently, a security architecture based on the assumption that society can simply produce sufficiently good citizens has an irreducible weakness. The dangerous actor of the future may not even have been a dangerous person yesterday.

The Problem Becomes More Serious as Technology Gets Smaller

Surveillance itself is also becoming more powerful.

In 2025, China’s National University of Defense Technology publicly demonstrated an insect-scale, mosquito-like micro-robot intended for reconnaissance and special battlefield missions. Public reporting should not be exaggerated into a claim that autonomous mosquito-sized surveillance fleets already exist: important questions remain about endurance, payload, communications, and operational deployment. Nevertheless, insect-scale robotics is real research, and National Defense University researchers have discussed the security implications of this class of micro-UAV.

The technological direction is important even if today’s prototypes remain limited.

Cameras once had to be mounted visibly on buildings. Future sensing systems may include tiny drones, distributed microphones, environmental sensors, autonomous robots, satellites, network telemetry, and devices that are already present in homes and workplaces.

This creates a disturbing technological symmetry:

The tools available to attackers become smaller, cheaper, and more autonomous—and so do the tools available to surveillance systems.

The contest therefore does not naturally end with less monitoring. It may end with considerably more.

China Is a Useful Example—but Not a Simple One

China is frequently associated with large-scale public-security video systems and facial recognition, which makes it an obvious reference point for this prediction. But the comparison should not be reduced to the caricature that China simply “watches everyone without rules.”

China introduced national regulations governing public-security video systems in 2025. Its government states that these systems collect, transmit, store, and process public-security video information while being subject to requirements concerning privacy and personal information.

China also introduced facial-recognition rules effective June 1, 2025. Among other things, they require necessity and proportionality, impose information-protection obligations, restrict making facial recognition the sole verification method when alternatives exist, and regulate facial-recognition equipment in public spaces.

The broader phenomenon is also not uniquely Chinese. AI-enabled surveillance technologies have spread internationally and are used by governments for a wide variety of purposes.

So the prediction is better expressed as:

The rest of the world may move toward technological capabilities that are today strongly associated with China, even while different countries retain radically different laws, political institutions, privacy protections, and limits on government power.

Mass Surveillance May Become a Security Equilibrium

Suppose a future biotechnology allows a small group to create an enormously dangerous pathogen.

A government has several options.

It can regulate laboratories and dangerous materials. It can secure DNA synthesis and other enabling infrastructure. It can improve intelligence. It can strengthen pandemic detection. It can develop medical countermeasures. It can monitor communications associated with credible threats. It can harden critical infrastructure.

Indeed, WHO biosecurity guidance emphasizes risk assessment, regulatory oversight, institutional safeguards, cybersecurity, and other preventive mechanisms—not universal surveillance as a single solution.

For this reason, it would be too strong to claim that mass surveillance has already been demonstrated to be the only possible defense against catastrophic technological misuse.

My prediction is instead political and technological.

As the maximum harm available to small actors increases, governments may repeatedly discover that every preventive mechanism still leaves gaps. Each catastrophic incident, attempted attack, or narrowly avoided disaster can then create pressure for one more layer of monitoring.

• More cameras.
• More transaction monitoring.
• More identity verification.
• More biological-material tracking.
• More network monitoring.
• More behavioral anomaly detection.
• More connections between previously separate databases.

The result can emerge incrementally without any government announcing: We have decided to create universal surveillance.

AI Changes the Equation Again

There is another reason this trend may accelerate: the dangerous actor no longer necessarily has to be human.

AI agents can plan, use software, communicate with external systems, operate computers, and perform sequences of actions with decreasing human supervision. NIST describes AI agents as systems capable of planning and taking autonomous actions that affect real-world systems or environments, while also warning that they introduce distinctive security challenges.

There is already evidence of AI being used much further along the operational chain than ordinary chatbot assistance. Anthropic reported in September 2026 that it had observed malicious cyber operations in which multi-agent systems performed reconnaissance, exploitation, and data exfiltration, sometimes operating for extended periods with limited human intervention. Anthropic stresses that humans still retained important decisions in many such cases, so this should not be interpreted as evidence of fully independent artificial terrorists. But the direction of travel is significant.

This produces a qualitatively different security problem.

Traditional surveillance asks:

What are people doing?

Future surveillance may also have to ask:

What are autonomous software systems doing?

AI Will Need to Watch AI

Humans cannot manually inspect every action performed by potentially billions of software agents.

That suggests an uncomfortable but technically straightforward conclusion:

Large-scale AI activity will probably have to be monitored primarily by other AI systems.

A defensive AI could continuously examine network traffic, autonomous-agent behavior, financial transactions, access requests, software execution, laboratory equipment, robotics, and other systems for patterns associated with dangerous activity.

That does not mean that the monitoring AI should possess unlimited authority. In fact, AI monitoring creates another security problem: who monitors the monitor?

Agents themselves can be attacked. Prompt injection, for example, attempts to manipulate an AI system through malicious instructions embedded in information the agent encounters. OpenAI describes modern prompt-injection defense partly as a problem of constraining the consequences of manipulation rather than assuming every malicious instruction can always be recognized in advance.

The same principle becomes much more consequential when AI systems monitor other AI systems.

From Human Surveillance to Machine Surveillance

We can therefore imagine several stages of technological development.

• First, humans watch humans.
• Then computers help humans watch humans.
• Then AI watches humans and alerts human authorities.
• Then AI watches both humans and other AI agents.

Eventually, there may be far too much activity for human beings to inspect directly. Human involvement shifts upward: establishing rules, auditing systems, investigating exceptional cases, authorizing particularly consequential interventions, and determining when automated systems have themselves become unreliable.

This is one reason AI safety is not merely a question of making one powerful model “nice.” It is increasingly a question of systems architecture: which machines are allowed to act, which machines observe those actions, which independent mechanisms detect abnormal behavior, and where human authority remains.

That broader problem also motivates Science DAO’s work on AGI safety. See the site’s discussion of the “Symbiote” AGI Safety Fund, which focuses on architectures for managing increasingly capable AI rather than assuming that a single aligned agent solves the entire problem.

Privacy May Survive, but in a Different Form

The strongest version of my prediction is not necessarily that every government employee will be able to inspect everyone’s private life.

Technology could produce a different model: pervasive machine observation combined with restricted human access.

For example, an automated system might continuously examine activity while retaining information cryptographically, exposing identifiable data to humans only after a narrowly defined risk threshold or judicial authorization is reached.

Whether such arrangements would actually protect civil liberties depends on their engineering, laws, governance, auditability, and enforcement. But conceptually, there is a substantial difference between:

“Nobody observes me,”

and

“Machines detect specified threats, but humans cannot freely inspect the underlying private data.”

Future privacy debates may increasingly concern the second problem because the first condition may become technologically and politically difficult to maintain.

The Security–Privacy Trade-Off May Become More Severe

Today’s debate often treats privacy and security as values that governments can rebalance through ordinary legislation.

Future technologies could make the trade-off considerably harsher.

If an individual can kill ten people, a society may tolerate a substantial probability that authorities fail to detect him.

If an individual can potentially kill ten million people, the politically acceptable probability of detection failure becomes dramatically smaller.

And if autonomous machines can attempt such actions thousands of times at machine speed, occasional surveillance becomes increasingly inadequate.

This is why I expect preventive monitoring to expand even in countries with strong civil-liberties traditions. Their implementations may differ substantially from China’s. Courts may impose stronger restrictions. Data may be decentralized. Monitoring algorithms may be externally audited. Warrants may remain necessary before humans obtain identities. Governments may prohibit particular forms of biometric tracking.

But the underlying technological direction could nevertheless be similar:

more of the world becomes observable.

This Is a Prediction, Not an Endorsement

There are strong objections to this future.

Surveillance systems can themselves be abused. Governments can redefine legitimate political opposition as a security threat. False positives can harm innocent people. Databases can be breached. Algorithms can discriminate or simply be wrong. Automated systems can magnify administrative mistakes across entire populations.

That means the argument “dangerous technology exists, therefore unlimited government surveillance is justified” does not follow.

Nor is surveillance the only countermeasure. Biosecurity, cybersecurity, controlled access to dangerous capabilities, resilient infrastructure, rapid outbreak detection, technical safeguards, privacy-preserving computation, and international coordination can all reduce risk.

My narrower prediction is that these measures may not eliminate the political demand for surveillance because offensive capability continues to diffuse.

The Future May Be a World in Which Everything Watches Everything

The long-term equilibrium could therefore be stranger than a conventional police state.

• Humans will watch machines.
• Machines will watch humans.
• Machines will watch other machines.
• Independent systems may watch the systems doing the watching.

A mosquito-sized robot is only a primitive symbol of that trajectory. The more important transformation is that sensing, computation, autonomy, and destructive capability are simultaneously becoming cheaper and more distributed.

The central question for the future may therefore stop being:

“Will we accept mass surveillance?”

It may become:

“Given that pervasive technological observation exists, who controls it, what is it permitted to detect, who can access the results, and who watches the watchers?”

I do not claim that this outcome is desirable.

I predict that increasingly powerful offensive technology will make some form of pervasive surveillance difficult for advanced societies to avoid. And because future threats will include autonomous software as well as human beings, AI will increasingly have to watch AI.

That may be one of the defining security architectures of the future.