You deserve the facts before you decide where you sit with AI.
Most of what you read is trying to scare you or sell you something. Neither is particularly helpful if you’re looking for agnostic, reliable information about the risk and reward of this technology.
So we went to the evidence. We wanted to get clear on what is known, what is still disputed, and what nobody can measure yet.
This report takes no side. Where the evidence is strong, it says so. Where a number is contested or comes from a single study, it says that too. Where informed people disagree, you get both positions.
You won't find a verdict on whether AI will cause the extinction of the human race, or if it will save us. What you will find is a sound basis for forming your own view, and for deciding where you sit on the comfort scale. And in turn, how this translate to adopting AI into your business.
The Loralai AI Risk Report
About the AI Risk Report
-
The report is broken into six key areas of concern.
How fast AI is developing. What the capability data shows, and where testing has fallen behind.
Extinction and loss of control. What experts estimate, and why nobody can measure it.
Environment. Energy, water and emissions, with the gaps in the data named.
Misuse. Fraud, deepfakes and cyber attacks that are happening now.
Privacy and data. What regulators and courts are enforcing.
Child safeguarding. Where the law stands, and where it has fallen behind the technology.
-
You can find a full, linked list of sources at the end of the report in the Appendix section.
Every source in the report also carries a grade, so you can see how much weight each claim will bear. It’s up to you to decide what sources you choose to trust.
Tier - What it Means
A - Independent, open about its methods, with no commercial stake. For example the International Energy Agency, Stanford HAI, the FBI and court records.
B- A company or regulator reporting on itself, under audit or regulatory pressure.
C - Commercial or advocacy research, or an estimate that rests on one method. Used only where nothing better exists, and always labelled.
-
It doesn't weigh these risks against the benefits of AI, which are also substantial and documented in many of the same sources.
It doesn't cover jobs, discrimination, mental health or misinformation. Those matter and deserve their own review.
It is a snapshot. Some of the events it cites were days old when we compiled it and may have moved on.
-
The report was produced using Claude (Opus 5.5), an AI model made by Anthropic.
We took great care with the original prompt and with every round of refinement. We asked for evidence over opinion, for every source to be graded, and for both sides to be set out wherever informed people disagree. That is how we kept the content factual, and how we guarded against the leanings of the tool itself.
It is also a good example of what we teach. Learning to prompt well is what turns AI chat tools such as Claude and ChatGPT into a source of relevant work with real depth.
Top Three Findings
AI is moving faster than our ability to measure and govern it.
Capability, investment and use are all growing faster than testing, regulation and oversight. Documented AI incidents rose 55% in a year, from 233 in 2024 to 362 in 2025. This is close to the only claim that every side of the debate agrees on.
The strongest evidence is about harm happening now.
Misuse and privacy have audited numbers, court records and named cases behind them. In 2025 the FBI recorded US$893.3 million in losses from fraud involving AI, the first time it has counted this as its own category. Fines under Europe's privacy law have passed €7.1 billion.
The biggest questions have the least data behind them.
Extinction risk is a serious position held by serious researchers. In the largest survey of AI researchers, the middle estimate was 5%. But nobody has ever observed the event, so every figure is a judgement, not a measurement. Environmental impact is real and growing, yet companies report water and emissions in ways that can't be compared. None of this makes the concerns wrong. It means the numbers alone can't settle them.
The Comfort Scale
Where do you sit on the comfort scale?
Some founders read this report and feel more at ease. Some feel less. Both are reasonable responses to the same evidence.
The research makes this point itself. On the hardest questions, thoughtful people look at the same facts and reach opposite conclusions.
Our job is not to move you along the scale. It is to make sure that wherever you land, you got there with good information.