2025 Predictions Revisited
Review of my 2025 predictions and recommendations for interesting 2026 predictions.
(For the recommendations on 2026 predictions, scroll waaayyy down.)
2025 predictions revisited
Last January1, I published my 2025 predictions. So how did I do?
Since I am grading my own work, I used Gemini, Perplexity and Claude to help review my predictions as objectively as possible2.
The problem with reviewing predictions is that everything you got right, seems obvious from today’s perspective. Everything you got wrong, now looks ridiculous or stupid. The beauty of hindsight!
What’s the verdict on my predictions overall? While I am not claiming any ‘super-forecaster’ skills, I think I can give myself a little pat on the shoulder with 9 ‘correct’ and 2 ‘mixed results’ ratings (but little time to rest on my laurels as I am working on publishing my 2026 predictions soon - subscribe to receive them in your inbox!).
Follow along for the details. The review provides a good summary of the developments of last year. And of course, please use the comment function generously to agree or disagree with my assessment.
Technology predictions
AI
“LLM progress will not slow down (high confidence). Here’s why I am confident on this point. While it seems that the approach of increasing compute and data is delivering diminishing returns, there are other avenues to continue the rapid pace of progress. Test-time compute / chain of thought seems to be particularly promising (e.g. OpenAI O1). Agentic AI systems are nascent but will be increasingly explored. Also, it will take companies years to even fully use the capabilities of existing models (see Narayanan / Ströbl / Kapoor article linked below). In short, LLMs will continue to keep us privacy and responsible AI folks very busy in 2025.”
✅ Correct prediction. At the end of 2024 / beginning of 2025, there were quite a few articles about generative AI plateauing or “hitting a wall”. With the pivot to “reasoning” models (based on inference technique / reinforcement learning from verifiable rewards - RLVR), generative AI has continued to improve along the same steep trajectory. Agentic AI was all the rage, but still faces significant issues for wide adoption as general AI assistants.
P.S. Simon Willison published a fantastic (albeit not short) summary of AI progress in 2025.
Blockchain / crypto
“ Still no good use cases in 2025 (medium confidence). Some people will likely make a lot of money (and conversely, a lot of people will lose a lot) from the speculation on cryptocurrencies under a likely more liberal regulatory regime in the US. However, outside of speculation and illicit activities, we will not see any further / different adoption of blockchain.”
❔Mixed results. While there have not been any great use cases for the average consumer, there has been a surge in stablecoin (as my generative AI assistants rightly pointed out). But spot on regarding the speculative and illicit activities (including the Trump family’s use of crypto).
Metaverse
“Still not ready for prime time (high confidence). The best way to understand my confidence in this prediction is to read Matthew Ball’s excellent book “Metaverse”. He lays out why we are still years away from having the infrastructure and technology required for the metaverse to work usefully.”
✅ Correct prediction. It’s not clear how long the metaverse is postponed, but it’s clearly not happening soon. Meta recently confirmed it lost 70bn on the metaverse, and Apple had weak demand for its VisionPro.
Quantum computing
“Will be one of the 2025 hypes but also bring the first interesting use cases (low confidence). This may be the year where quantum computing moves from experimental technology to establishing the first (limited) business use cases.”
❔Mixed results. I originally intended to give myself bad marks for this one, but Gemini and Claude both considered this correct given proof-of-concepts (POCs) in drug discovery and battery logistics.
Privacy regulation / policy predictions
US legislation
“Still no overarching (8) federal privacy law with states continuing to fill the gap (high confidence). There is hope that privacy could be one of the common, bipartisan policy agenda items under the Trump administration / Congress. This may be true about the general direction, but I think the usual two contentious points, private right of action and preemption of state laws, will get in the way … again. More state privacy laws, but no federal privacy law. I would be thrilled to be wrong, as the US desperately needs an overarching federal (consumer) privacy law.”
✅ Correct prediction. But let’s be honest, that’s a pretty safe bet to make (likely also for next year).
US regulation
“A more business-friendly and politicised FTC, but no significant change in privacy enforcement (medium confidence). With new chair Ferguson, there will be a shift in tone (no more “commercial surveillance”) but quite possibly no significant difference in privacy enforcement actions taken based on Ferguson’s record.”
✅ Broadly correct prediction. However, I did not expect that the FTC would set aside an AI-related consent order. I also didn’t make it explicit that I was only referring to federal regulation. State regulators (e.g. CPPA) were, as expected, not necessarily following the federal direction.
EU data transfers
“Data transfers will be back in focus, but the DPF will survive (high confidence). The new US administration may be unhappy about the constraints which Executive Order 14086 (issued by the Biden administration to enable the EU-US Data Privacy Framework) places on US security agencies. Or the Trump administration may just revoke it in the context of a more adversarial relationship with the EU. In any case, the fate of EO 14086 seems uncertain. There may also be other Trump administration changes that impact the EU Commission’s (and the DPAs’) assessment of the equivalence of the US system regarding protecting transferred personal data. The EU Commission will be keen to keep the DPF alive despite any US actions, but there will likely be challenges by NGOs or DPAs should EO 14086 be revoked.”
✅ Correct prediction. EO 14086 wasn’t revoked. The EU General Court upheld the adequacy decision in September 2025 (‘Latombe’ decision). But with the unstable EU - US relationship, the DPF has been - and will remain - vulnerable.
EU GDPR
“The GDPR will be updated in the next two years to ensure LLMs can be developed legally with efforts starting in 2025 (medium confidence). The EDPB and DPAs are currently trying their best to square the circle, but the question of legal basis for LLM training and development (and for the subsequent use) can only be fudged for so long. The bullet on the GDPR in von der Leyen’s Mission letter (page 7) to EU Justice Commissary McGrath seems to hint at GDPR changes.”
✅ Correct prediction … and probably my proudest one. I am normally pretty humble, but couldn’t resist taking a little victory lap on LinkedIn when it was confirmed to be part of the Digital Omnibus package. Except for that small bullet on the mission letter, there was very little indication of this actually happening in 2025. Geopolitics is the gravity of tech regulation.
Privacy regulation / policy predictions
US
“No US federal AI law and looser regulation on federal level will lead to a patchwork of state laws (high confidence). There are very few signs that the US will legislate an equivalent to the EU AI Act. Quite the opposite, Trump has vowed to repeal Executive Order 14110 on AI. Under the Trump administration, AI policy and regulation is likely to be more business-friendly, less globally coordinated and more focused on US national security interests. Similarly to privacy, US states will follow the example of Colorado and legislate in the absence of a federal law. Despite a published commitment to use model laws / alignment around core principles (see open letter of the Multistate AI Policymaker Working Group), we will likely end up with a patchwork of state laws that follow slightly different models (like in other legal areas). On the federal level (and globally), the NIST AI Risk Management Framework will remain unscathed and influential.”
✅ Broadly correct prediction. I made this prediction before it took office, but the Trump administration has indeed gone down the deregulation / AI race with China route. However, US states have not been as proactive with regulation as I expected - likely also due to the Trump administration’s efforts to impose a moratorium on state laws (attempt to include it in the OBBBA and with an executive order in December). I also did not anticipate that NIST would suffer from budget cuts and lose key staff.
EU
“The EU will come to the conclusion that it over-regulated AI (and missed some risks) and will try to update the EU AI Act through the implementing standards and guidelines (medium confidence). 2025 will be the year when both companies and the multitude of EU and Member State authorities (see hair-raising governance chart by Future of Privacy Forum) will struggle to implement the EU AI Act. The stringent, product safety-based requirements will be a significant challenge for start-ups and SMEs. The EU AI Board, the EU AI Office and the many Member State authorities will face pressure to implement the EU AI pragmatically whilst trying to get organised themselves. At the same time, there will be efforts to address AI risks which the EU AI Act has not anticipated (e.g., AI companions).”
✅ Broadly correct prediction. While the EU insists that the EU AI Act approach is generally sound (a mistake, I think), it has reopened the EU AI Act with the Digital Omnibus package - including the aim to centralise more powers in the EU AI Office and delay high-risk requirements. However, there haven’t been any attempts to better address AI companion risks and other risks not included in the high-risk categories (will this possibly be included as part of the Digital Omnibus negotiations this year?).
Global
“No “Brussels effect” for AI regulation (medium confidence). While many countries will follow the implementation of the EU AI Act with interest, I suspect that many countries will try to learn from the EU’s mistake and regulate AI more flexibly and pragmatically rather than taking a full-on product-safety-approach like the EU (see bullet above). Sure, most countries will also apply a risk-based approach with different requirements depending on the risk of AI systems. However, I think only few will copy the EU AI Act in a way that GDPR inspired laws around the world. Predicting this for 2025 is not a big gamble, since it’s still early days. But I predict the same will apply for the following years.”
✅ Correct prediction. The Trump administration, of course, is very much on collision course with the direction of the EU AI Act. But also other countries like Japan, South Korea and India have taken more innovation-friendly and less stringent approaches. Other countries like the UK, have been relying on existing laws and regulatory guidance and adopting a ‘wait-and-see’ approach. Given that the EU itself reopened the EU AI Act through the Digital Omnibus package, the same prediction will likely be true for next year.
Recommended 2026 predictions
Phew, that was a lot to get through. As a little reward, here are some interesting articles with predictions for this year that I have used to start gathering my 2026 predictions. Let me know if you know of any other good articles!
The Economist - The World Ahead 2026 (£): https://www.economist.com/topics/the-world-ahead-2026 (includes a few interesting articles on AI - see my short Bluesky thread for some highlights)
The FT - Business trends, wild cards and companies to watch in 2026 (£): https://www.ft.com/content/c2832e0f-d858-4479-b341-def386e5cf0d?shareType=nongift
Scott Galloway (No Mercy / No Malice) - 2026 Predictions: https://www.profgalloway.com/2026-predictions/
Techradar: 5 predictions for AI in 2026 – what’s next for ChatGPT, Gemini, and you: https://www.techradar.com/ai-platforms-assistants/5-predictions-for-ai-in-2026-whats-next-for-chatgpt-gemini-and-you
FT - Why we should know what we don’t know (insightful article on our biases and how they impact predictions) (£) https://www.ft.com/content/d8b26855-c6d8-4d76-b9f2-663e90548d6c?shareType=nongift
Paul Fairie - A List of Predictions Made in 1926 About 2026:
The Guardian / Stephen Collins - My predictions for 2026, by Darren the dog: the Stephen Collins cartoon: https://www.theguardian.com/lifeandstyle/picture/2026/jan/02/my-predictions-for-2026-by-darren-the-dog-the-stephen-collins-cartoon
2 January 2025, to be precise (so before the Trump administration took over).
My AI research assistants generally performed well, but both Claude and Perplexity struggled with the Digital Omnibus-related prediction.