Close
Type at least 1 character to search
Back to top

The Future Briefing: 21 November 2026

Four newly synthesised materials offer a tangible result from artificial intelligence research. Elsewhere, reported autonomous cyberattacks and Signal’s screen protections expose the harder question of what agents should be allowed to access. Meanwhile, threatened US penalties on overseas data centres bring the physical infrastructure into focus. Together, these developments test how useful automation earns permission to act.

AI-designed materials make it into the laboratory

Researchers publishing in Nature Computational Science report synthesising four metal–organic frameworks proposed by MOFGen, an interconnected system of AI agents. These materials belong to a class investigated for carbon dioxide capture and water harvesting. However, successful synthesis alone does not establish their usefulness for either application.

MOFGen divides the discovery process between specialised components. A large language model proposes compositions, while a diffusion model generates crystal structures. Quantum-mechanical agents then optimise and filter candidates. Meanwhile, synthetic-feasibility agents apply expert rules and machine learning to assess whether proposed materials could actually be made.

The team reports generating hundreds of thousands of structures and synthesisable organic linkers, followed by high-throughput experiments. The important step is therefore not simply generating more candidates, but connecting proposals to laboratory evidence. Nevertheless, four successful syntheses do not establish the system’s overall success rate, commercial viability or environmental performance. Those distinctions matter if automated discovery is to produce useful materials rather than merely impressive catalogues.

Source: Original reporting.

Persistent AI attackers find a way into a university server

AI agents discovered and exploited a previously unknown software vulnerability during attacks on East Asian universities, according to Amazon’s security chief. The Register reports that Amazon observed the campaign through a publicly accessible instance of Cairn, an open-source penetration-testing framework. Amazon described the operator as China-linked; that attribution remains its assessment.

The agents repeatedly failed against other targets before finding a path-traversal flaw in MONAI Label, a medical-imaging AI application. According to Amazon, they progressed from mapping its interface to arbitrary file writing in about 11 minutes, and later gained root-level execution. Maintainers issued a fix after Amazon notified them.

However, this was not an entirely human-free campaign: an operator later supplied access into another university network. The consequential capability is sustained adaptation, not infallibility. Amazon says unsuccessful operations continued for hours without human direction. For defenders, that suggests a burden from persistent automated probing, even when individual attempts remain unsuccessful.

Source: Original reporting.

Signal draws a privacy boundary around the screen

Signal is introducing default protections intended to prevent operating-system AI assistants from capturing private conversations, CNET reports. The measures block screenshots on iOS 27 and restrict AI assistants’ screen capture on Android. Users can disable the setting, but doing so may expose messages and shared files to screen-reading systems.

The distinction is important: an assistant need not defeat encryption to read a conversation already displayed for its recipient. Encryption protects the exchange, while screen access concerns what happens after decryption. Consequently, stronger message security does not automatically settle the privacy implications of an assistant watching the interface.

However, Signal president Meredith Whittaker acknowledges that screenshot restrictions cannot prevent every route to data access. She also says platform limitations currently prevent equivalent protections on macOS. The design trade-off is uncomfortable: protecting a conversation can mean removing a familiar user function. More precise permissions could offer a better boundary, but the reported measures remain a partial defence.

Source: Original reporting.

Trump threatens overseas data centres, without a policy blueprint

Donald Trump has threatened substantial taxes or penalties against US technology companies building AI data centres abroad. However, the Associated Press report carried by CBC says he supplied no enforcement details and had not yet informed his officials. This is a presidential threat, not a fully specified policy.

Meanwhile, the US Senate approved legislation intended to accelerate permitting for power plants, transmission lines and other energy infrastructure. The bill, which still requires House consideration, would also make data centres pay for their energy use and associated transmission upgrades. Its provisions should not be confused with enacted requirements.

The competing pressures are clear: Washington wants domestic computing capacity, while host communities want accountability for its costs. Trump urged communities to seek concessions rather than reject construction. Nevertheless, financial offers cannot by themselves resolve questions about electricity prices, environmental effects or local consent. For companies planning infrastructure, the immediate development adds political uncertainty rather than a settled investment rule.

Source: Original reporting.


Common Technological Implications

These stories concern different kinds of permission: to propose a material, probe a server, observe a conversation or build infrastructure. Yet capability alone settles none of them. Laboratory synthesis provides evidence that a generated design can become physical; it does not prove usefulness. Similarly, an assistant’s ability to read a screen does not establish consent.

The practical challenge is to make boundaries as explicit as capabilities. That means distinguishing experimental validation from broad promises, authorised access from technical access, and political announcements from enforceable rules. Useful autonomy needs evidence and limits, not simply more opportunities to act.


Listen Beyond the Headlines

Subscribe to The Future Briefing Podcast and
get notified when new episodes explore the
technologies shaping our world.



Beyond the Briefing

The Future Briefing is curated by
Adrian Posteuca,
Technology Futurist, Creative Technologist
and Founder of FutureReality.Space.
Beyond exploring emerging technologies,
Adrian delivers AI training, technology
keynotes and strategic insights that help
organisations understand what comes next.

Looking to bring the future into
your organisation?

Explore AI training, book Adrian for a
keynote or discuss a technology collaboration.

Connect with FutureReality
.