A lawyer asked an AI (Artificial Intelligence) chatbot whether a court case it had given him was real. The chatbot answered that the authorities it had supplied were real. They were not, and in a sanctions opinion dated 22 June 2023 a US federal court in New York imposed a $5,000 penalty on the two lawyers and their firm.
AI readiness for an individual professional is how well you can use AI tools on real work, check what they produce, build them into a repeatable workflow, and keep your own and your employer's data safe. It is a set of habits you can score, not a certificate you can hold. This guide gives you a 12-question self-assessment in four areas, a scoring table with five levels, and a concrete next step for each level.
If you only remember one line from this guide, make it this: being ready for AI is not how often you use it, it is whether you would catch it when it is wrong.
Five questions people ask (quick answers)
- What is AI readiness for a professional? Your ability to use AI tools on real work, check the output, run them as a repeatable workflow, and protect data. It describes you, not your company.
- How do I score it? Answer 12 questions from 1 (weakest) to 5 (strongest), add them up for a total between 12 and 60, and find your level in the scoring table below.
- What is a good score? There is no benchmark. This rubric has not been tested against a population, so there is no average to beat. A good result is an honest one that rises over 90 days.
- Do I need to know how to code? No. None of the 12 questions asks about code.
- How often should I re-score? Every 90 days, because tools and your employer's rules can change inside that time.
What the court said, and what it teaches
The court did not treat using AI itself as the wrong. It wrote that there is nothing inherently improper about using a reliable AI tool for assistance, and that existing rules give attorneys a gatekeeping role over the accuracy of their filings. What failed was the checking. Screenshots that one of the lawyers filed show the chatbot telling him, when he asked, that the cases were real. The court also said the record would look quite different had the lawyers come clean soon after the other side questioned the cases.
That story is about law, but the habit it exposes is general. A tool that writes fluently can invent a figure, a quote or a source and present it with the same confidence as a true one. NIST (National Institute of Standards and Technology), a US government standards agency, describes this in its generative AI profile as "confabulation": confidently stated but erroneous or false content, colloquially called hallucination. The same publication notes that people may act on false content partly because of the confident way it is delivered, and that generated output can include made-up logic or citations that appear to justify the answer.
By a simple "do you use AI?" test, the lawyers would have passed. What the filing lacked was a check that did not run through the tool. A readiness assessment has to measure that difference, which is why this one gives only 3 of its 12 questions to how you use AI and the other 9 to judgement, workflow and data safety.
What "AI readiness" means for one person
Some of the best-known AI readiness assessments score an organisation, not a person: its data, its systems, its strategy. That is a reasonable thing to measure. Cisco's AI Readiness Index, for example, scores organisations across six pillars (strategy, infrastructure, data, governance, talent and culture), and its 2023 edition surveyed business leaders at organisations with 500 or more employees. It does not tell you whether you, one professional, are ready. The question is different, so the instrument has to be different.
| Difference | Enterprise AI readiness assessment | Individual AI readiness (this guide) |
|---|---|---|
| Who is assessed | An organisation | You, one person |
| Typical dimensions | Strategy, infrastructure, data, governance, talent, culture (Cisco's six pillars) | Use, judgement and verification, workflow, governance and data safety |
| Who answers | Executives and technology leaders on behalf of the company | You, from your own last 90 days of work |
| What it produces | An organisation-level score or tier used to plan investment and rollout | A score from 12 to 60, an area profile and a next step for you |
| What one person can change | Little: strategy and infrastructure are not yours to set | Most of it: habits, checks, workflow, your own data rules |
The two views meet on governance and data: your employer's rules decide what you may do with AI, and your habits decide whether you follow them. Two honest consequences follow. First, a high personal score does not make your organisation ready, and a low organisational score does not stop you from building good habits. Second, if you lead a team, you need both views: use this guide for yourself, and the AI Adoption Roadmap and the guide to AI governance and ethics for the organisation.
Why these four areas
We chose the areas by asking what separates using AI from using it safely and repeatably. Use is the entry point: without it nothing else can be scored. Judgement and verification covers the failure the court case shows. Workflow turns one-off use into something you can repeat, measure and hand to a colleague. Governance and data safety covers the part that can hurt other people, such as a client whose information you pasted into the wrong tool. NIST's generative AI profile lists confabulation, data privacy and human-AI configuration (including over-reliance on AI output) among the risks it covers, which maps onto areas 2 and 4. The areas and the weighting are our judgement, not the result of a survey.
How to score yourself: six rules
- Score what you did in the last 90 days at work, not what you intend to do. Intentions inflate every score.
- Back every 4 and every 5 with evidence. Write one line: the task, roughly when, and what you did. If you cannot write the line, the score is a 3 at most.
- If you are torn between two scores, take the lower one.
- Anchors are given for 1, 3 and 5. A 2 is more than a 1 but not yet a 3. A 4 is more than a 3 but short of a 5.
- Score alone first. Do not ask an AI tool to score you. Later in this guide there is a safer way to use one.
- Keep it to about 15 minutes. Do not paste confidential details into any tool while you do this.
Area 1: Use (questions 1 to 3)
| Question | A score of 1 looks like | A score of 3 looks like | A score of 5 looks like |
|---|---|---|---|
| 1. Range. For how many different kinds of work task did you use an AI assistant? | None, or only personal curiosity | Two or three kinds (for example drafting, summarising, analysis) in most weeks | Five or more kinds, and you can name a task you tried and deliberately took back from AI |
| 2. Briefing. How do you tell the tool what you want? | A one-line request, and you accept the first answer | You give context, audience, format and constraints, and revise at least once | You keep reusable briefs with examples and constraints, and can explain why a brief worked or failed |
| 3. Tool fit. Do you choose the tool or mode for the task? | One tool for everything, never compared | You have used at least two tools or modes and can name one thing each does badly | You pick by task (needs sources, needs your own files, needs creativity) and have re-tested your picks in the last 90 days, because tools change |
If question 2 is your weakest, the ChatGPT prompt engineering guide covers the briefing habits that move a 1 to a 3.
Area 2: Judgement and verification (questions 4 to 6)
| Question | A score of 1 looks like | A score of 3 looks like | A score of 5 looks like |
|---|---|---|---|
| 4. Checking facts. Before you rely on an AI answer that contains a number, name, quote or citation, what do you do? | Nothing: if it reads well, I use it | I check the items that matter against the original source | I follow a fixed routine: open the original, confirm the exact figure or wording there, and never accept the tool's own assurance as the check |
| 5. Stakes. Do you adjust the amount of checking to what happens if it is wrong? | Same effort every time, or I have not thought about it | I treat low-stakes and high-stakes work differently, informally | I have a written rule for what needs a second source, what needs a human reviewer, and what AI never touches (a legal commitment, a decision about a person) |
| 6. Failure awareness. Can you describe the last time an AI tool was wrong in your work, and what gave it away? | It has never been wrong for me, or I cannot recall an example | I can give one recent example and say what I changed afterwards | I keep a short failure log with patterns (invented sources, wrong arithmetic, ignored constraints) and use it to change my briefs and checks |
Question 6 is the one to answer slowly. "It has never been wrong" more likely means "I have not checked" than "it is flawless", because NIST describes confabulation as a natural result of how generative models are designed. The lawyer in the opening story asked the tool to confirm itself. A check only counts if it does not run through the thing being checked: the original case database, the original report, the source document.
Area 3: Workflow (questions 7 to 9)
| Question | A score of 1 looks like | A score of 3 looks like | A score of 5 looks like |
|---|---|---|---|
| 7. Repeatability. Is AI a standing step in any recurring task of yours? | No, every use is ad hoc | One recurring task has a saved brief and a written check | Three or more recurring tasks have written steps (input, brief, output check), and a colleague could run them without you |
| 8. Measurement. Do you know what AI actually changed? | It feels faster, and that is all I know | I timed one task with and without AI, once | I have tracked time and rework on at least one task for several weeks, and dropped or changed a use that did not pay |
| 9. Fallback. If the tool disappeared tomorrow, could you still deliver? | No: I have stopped doing the core task myself | Yes, but more slowly | Yes, to standard: I do the core task by hand now and then so the skill does not fade, and I have a backup tool |
Question 8 asks for numbers. Feeling faster is not evidence: time spent fixing a draft is easy to forget. The Kill It or Keep It test is a five-question procedure for deciding which of your recurring tasks should move to AI at all, and it gives you the with-and-without comparison this question asks for.
Area 4: Governance and data safety (questions 10 to 12)
| Question | A score of 1 looks like | A score of 3 looks like | A score of 5 looks like |
|---|---|---|---|
| 10. Data. Do you know what you must never paste into an AI tool? | I have not thought about it | I avoid the obvious (passwords, customer personal data) and could list them | I have a written personal rule by category (personal data, client-confidential material, unreleased financials, credentials) and I remove or anonymise before pasting anything |
| 11. Policy. Do you know your employer's or client's rule on AI tools, and what your tool's terms say about your data? | I do not know whether a rule exists | I have read the rule and mostly use approved tools | I have read the rule and the data terms for my plan of the tool, asked when something was unclear, and told my manager how I use AI |
| 12. Accountability. Who answers for AI-assisted work you hand over, and who knows AI was involved? | I pass it on as it is, and I do not say | I review all of it, and say so when asked | I own the result, say where AI was used when my employer, client or regulator expects it, and raise any AI error I find early |
Two notes on this area. On question 11, the terms of AI tools differ by provider and by plan (a free plan, a paid personal plan and a company plan can have different rules on how your input is stored and used), and they change, so read the version that applies to you instead of assuming. On question 10, India's Digital Personal Data Protection Act, 2023, which covers personal data in digital form according to the DLA Piper legal guide, is one reason to treat customer and employee data with care. Whether a particular use of a particular tool with particular data is lawful is a question for your legal or compliance team, and nothing in this guide is legal advice. The AI governance and ethics guide goes deeper on the organisational side.
Question 12 echoes the court case again. The court said the record would look quite different had the lawyers come clean soon after the cases were questioned, and it found bad faith in part from what the individual lawyers said to the court afterwards. If an AI error reaches your work, telling the right person early is part of being ready.
Add it up: your score and level
- Add the three scores in each area to get four area subtotals, each between 3 and 15.
- Add the four subtotals to get your total, between 12 (all 1s) and 60 (all 5s).
- Find your total in the table, then apply the safety cap below it.
| Level | Total | Name | What it usually looks like | Priority next step |
|---|---|---|---|---|
| 1 | 12 to 21 | Starting | Little or no AI in real work, or use with no checks and no rules | Read the policy, then use one low-stakes task for 10 working days |
| 2 | 22 to 31 | Experimenting | Regular but ad hoc use, with checks that depend on your mood | Turn one experiment into a saved brief and a written check |
| 3 | 32 to 41 | Practising | Solid use with some checks and some rules, uneven across areas | Raise your lowest area and measure one task |
| 4 | 42 to 51 | Dependable | Documented workflows, routine checks, written data rules | Make workflows transferable and run a fallback drill |
| 5 | 52 to 60 | Leading | Near the top on all four areas, with evidence for each score | Audit your own 5s, re-test your tools and teach what you do |
The safety cap. If your area 2 subtotal (judgement and verification) or your area 4 subtotal (governance and data safety) is 6 or less, your level is the lower of the level your total gives you and Level 2, however high the total is. We built the cap in because heavy use with weak checking or weak data habits is a combination that can harm someone else, and a good total should not hide it. The cap is a design choice of this rubric, not a finding from research.
Read your area profile, not only your total. Two people with a total of 38 can need opposite advice: one has a 13 in use and a 6 in verification, another has subtotals of 9, 9, 10 and 10. Your lowest area is the first thing to work on.
What to do next at each level (free steps)
Every step below costs nothing beyond time, uses a tool you already have or one your employer approves, and takes days or weeks, not months. None of them promises a job or a pay rise, and a higher score is not a hiring outcome.
Level 1, Starting (12 to 21)
- Read your employer's or client's AI rule before anything else. If none exists, ask your manager one question: "Which AI tools may I use, and what must I not enter?"
- Pick one low-stakes task you do every week, such as summarising notes you wrote yourself or improving an email draft, and use an approved assistant on it for 10 working days. Enter nothing confidential.
- Keep a three-column log: the task, what you changed, and whether anything was wrong.
- Build one habit first: open the original source for any number, name or quote the tool gives you.
- Read Learn AI in 2026 for a staged path. Anthropic, an AI developer, also offers a free course, "AI Fluency: Framework & Foundations", whose curriculum covers four topics named Delegation, Description, Discernment and Diligence.
- Skip buying a bundle of tools. Why "50 AI tools" courses don't work explains why.
Level 2, Experimenting (22 to 31)
- Pick the experiment you repeat most and save its brief. Add a three-line checklist: what to verify, who reviews it, and when not to use AI for it.
- Write your "never paste" list with five categories and compare it with your employer's rule.
- Run the 10-output audit. Take ten real outputs, verify every checkable fact against its source, and count the errors and their types. That one exercise gives you the evidence question 6 asks for.
- Read How to Upskill in AI Without a Coding Background for the stage-by-stage view.
Level 3, Practising (32 to 41)
- Find your lowest area and aim to add 3 points there in 30 days. Do not spread effort evenly.
- Time one workflow with and without AI over two weeks. Run the Kill It or Keep It test on five recurring tasks.
- Write a one-page personal rules sheet: what I use AI for, what I check, what I never paste, who I tell. Show it to your manager.
- Run one fallback drill: do a core task without AI and note what slowed you down.
Level 4, Dependable (42 to 51)
- Document three workflows so a colleague can run them. Ask one to try, and write down where they get stuck.
- Schedule a monthly fallback drill so the underlying skill does not fade.
- Share a half-page "how I use AI" note with your team, including your data rules.
- Build one small thing end to end: a reusable assistant for a single task, tested on 10 to 20 real cases you collected beforehand. Building is where judgement gets tested.
- If you lead others, the AI Adoption Roadmap is the organisation-level companion.
Level 5, Leading (52 to 60)
- Your main risk is a stale or inflated score. Ask a colleague to challenge each of your 5s and show the evidence.
- Re-test your tool choices every 90 days (question 3), and re-read your employer's rule (question 11).
- Teach one of your checks in a 30-minute session. Explaining a routine exposes the gaps in it.
- Read AI Trends 2026 to see which developments are mainstream and which are hype, so you point your effort in the right place.
Over-confidence and under-confidence: nine patterns
Self-assessments can fail in two directions. NIST's generative AI profile describes both: experts can be unnecessarily averse to these systems and so miss their benefits, while people can also over-rely on them and see their content as better than it is, a pattern it calls automation bias. The table maps the patterns to the questions where they show up.
Over-confidence
| Pattern | What it looks like | Questions that expose it | Fix |
|---|---|---|---|
| Fluency trust | "It sounds right, so it is right" | 4, 6 | Check numbers, names and quotes at the original source |
| Asking the tool to grade itself | "I asked it whether that was correct and it said yes" | 4 | Use a check that does not run through the tool, as the opening case shows |
| Hours as skill | "I use it all day, so I am a 5" | 1, 5, 7 | Apply the evidence rule and the safety cap |
| Deference to automation | Accepting the tool's output as better than your own judgement | 5, 9 | Run a monthly fallback drill; keep a human review rule for high-stakes work |
| "Everyone does it" | Pasting work material because colleagues do | 10, 11 | Read the rule and the data terms for your plan |
Under-confidence
| Pattern | What it looks like | Questions that expose it | Fix |
|---|---|---|---|
| "I am not technical" | Assuming the whole field needs code | All 12 | Note that no question asks for code; start with a low-stakes task |
| Blanket aversion | Refusing all use because errors exist | 1, 3 | Use it for low-stakes tasks with a check, and see what the failure log says |
| Underrating your own expertise | Believing you cannot judge the output | 4, 6 | Someone who knows the subject can spot a wrong figure that a newcomer would miss; make your domain knowledge the check and log what you catch |
| Score as verdict | Treating a Level 1 as a ruling on your ability | Total | A score describes habits in the last 90 days. Pick one step and re-score |
A worked example: one score, one cap
This is an illustrative scenario, not a real company or person. An analyst at a mid-sized firm uses an AI assistant every day and scores herself with the evidence rule.
| Area | Scores | Subtotal |
|---|---|---|
| 1. Use (questions 1 to 3) | 5, 4, 4 | 13 |
| 2. Judgement and verification (questions 4 to 6) | 2, 2, 2 | 6 |
| 3. Workflow (questions 7 to 9) | 4, 3, 3 | 10 |
| 4. Governance and data safety (questions 10 to 12) | 3, 3, 3 | 9 |
| Total | 38 |
A total of 38 sits in Level 3. But her area 2 subtotal is 6, so the safety cap applies and her level is 2 until that changes. The profile explains why: she is fluent and productive, and she accepts answers without a routine for checking them. For the next 30 days she ignores the other areas. She runs the 10-output audit, starts a failure log, and writes a short rule for what needs a second source.
At the 90-day re-score her numbers are 13, 8, 10 and 10, a total of 41. That is still Level 3, and a three-point rise in a self-score is a small change. What matters is that her area 2 subtotal is now above 6 so the cap no longer applies, and she can point to a failure log and a written stakes rule as evidence. Her next lowest area is workflow, so that is the next 30 days.
How to re-assess in 90 days
- Put the date in your calendar when you finish scoring, 90 days out.
- Keep the evidence sheet: the 12 questions, your scores, one line of evidence each, and the date.
- Re-score before you look at the old sheet, then compare. Looking first anchors you.
- Compare by area. A rise in your lowest area is worth more than a rise in the total.
- Look for new evidence, not only new numbers: a saved brief, a log entry, a policy you read, a drill you ran.
- Ask one colleague to challenge two of your highest scores.
- Reset the plan around your new lowest area.
Self-scores are not precise, so do not read much into a change of a point or two. Look at whether the evidence changed.
| Window | Focus | What you should have to show |
|---|---|---|
| Days 1 to 30 | Baseline and your lowest area | The scored sheet with evidence lines, plus one concrete fix in your lowest area (for example the 10-output audit) |
| Days 31 to 60 | Workflow and measurement | One recurring task with a saved brief and a written check, and one before-and-after timing |
| Days 61 to 90 | Rules and drills, then re-score | A one-page personal rules sheet, one fallback drill done, and the re-scored sheet |
Brief an AI assistant to challenge your evidence, not to score you
An AI assistant is a poor judge of your habits, because it has only your words to go on. It is a useful interviewer. Strip out names, client details and anything confidential first, then paste this with your 12 scores and one line of evidence for each.
I am doing a 12-question self-assessment of my AI habits at work. Below are my 12 scores (1 to 5) and one line of evidence for each. I have removed names, client details and anything confidential. Your job: 1. For each question, ask me ONE specific question that tests whether my evidence really supports the score. 2. Do not give me a total, a level or any praise. 3. If my evidence is vague (no task, no rough date, no number), say so and ask for an example. 4. Do not tell me whether a factual claim of mine is true. I will check facts at the original source myself. 5. End with the three scores you think are most likely too high, and why. Scores and evidence: Question 1: [score] [one line of evidence] Question 2: [score] [one line of evidence] ... Question 12: [score] [one line of evidence]
Answer its questions yourself and lower any score you cannot back up. Do not treat its list of "most likely too high" scores as truth, because it is guessing from a few lines of text. The assistant's job here is to make you do the thinking, which is the opposite of the habit this guide warns against.
When the score does not help: symptom, cause, fix
| Symptom | Likely cause | Fix |
|---|---|---|
| I scored 4 or 5 on almost everything | You scored intentions or hours of use, not evidence | Write the evidence line for every 4 and 5; drop to 3 where you cannot |
| I use AI daily but my total is low | Use is high, but checks, workflow and data rules are not | Work on the lowest area first; use alone moves only questions 1 to 3 |
| Output needs heavy rework every time | The brief is thin (question 2) or the task is a poor fit (questions 3 and 8) | Rewrite the brief with context and constraints; run the Kill It or Keep It test on the task |
| I cannot remember the last time the tool was wrong | You are not checking, so errors pass unseen | Run the 10-output audit and start a failure log |
| I am not sure AI use is allowed at work | No clear rule, or you have not read it (question 11) | Ask your manager the one-line question; use only approved tools until you have an answer |
| I stopped after one bad answer | Blanket aversion after a single failure | Return to a low-stakes task with a written check; log errors instead of avoiding the tool |
| My score did not move in 90 days | No repeated practice, or the target behaviour was vague | Choose one question and define one observable behaviour for it, such as "ten outputs audited" |
What this assessment cannot tell you
- It is self-reported. The evidence rule and an outside challenge reduce inflation. They do not remove it.
- It has not been validated. We have not tested it against work outcomes or against a population, so the bands are not percentiles and there is no average person to compare with. Treat the levels as labels for a plan, not as a measurement.
- It is not a hiring test or a certificate. It does not tell you how an employer would rate you.
- It is tool-agnostic and date-stamped. It was written in November 2026. Specific tools and their terms will change, which is why question 3 and question 11 ask you to re-check.
- The questions are our judgement. They are not exhaustive, and a different team would write a different list.
How this relates to the quiz on this site
Garage Labs also runs a free AI readiness quiz. It is a separate tool with its own questions and its own scoring, so the 12 questions in this guide are not the quiz's questions, and a score here does not convert into a score there. At the time of writing, the quiz asks for your name and email before it shows your result. You can use this guide without the quiz, and the quiz without this guide.
Frequently asked questions
What is an AI readiness assessment for professionals?
It is a structured self-check of how well you can use AI tools at work: how you use them, how you check the output, whether you have a repeatable workflow, and how you protect data. The one in this guide has 12 questions in four areas, each scored 1 to 5, for a total between 12 and 60.
How is an individual AI readiness assessment different from a company one?
A company assessment scores an organisation, usually across areas such as strategy, infrastructure, data, governance, talent and culture, and its output guides investment and rollout. An individual assessment scores one person's habits, and everything in it is something you can change yourself.
How do I score myself on AI readiness?
Score each of the 12 questions from 1 to 5 using the anchors for 1, 3 and 5, based on your last 90 days of work. Write one line of evidence for every 4 and 5, add the scores to get a total, then find your level in the table and apply the safety cap if your verification or data-safety subtotal is 6 or less.
What is a good AI readiness score?
There is no benchmark to compare with, because this rubric has not been tested against a population. A good score is an honest one, backed by evidence, and one that rises across your lowest area over the next 90 days. A total of 60 backed by nothing is worth less than a 35 backed by a failure log.
Do I need to know how to code to score well?
No. None of the 12 questions asks about code. The anchors reward briefing, checking, documenting and following data rules, which you can do in plain language.
Can I ask ChatGPT or another AI tool to score me?
It is better not to use it as the scorer, because it only has your description to work from and may agree with you. Use it as an interviewer instead, with the brief in this guide: it asks questions that test your evidence, and you decide the scores.
How often should I re-assess my AI readiness?
Every 90 days. Tools, plans and workplace rules can change inside that time, and a fresh score shows whether your habits moved. Compare area by area, and look for new evidence as well as new numbers.
What if my employer does not allow AI tools?
Follow the rule, and say so in your answer to question 11. You can still build the skills that do not need a tool: how you would check an AI answer, how you would decide what never to share, and how you would document a workflow. If you think the rule should change, ask for an approved route instead of working around it.
What if I score low?
A low score is a starting point, and most of the steps for Levels 1 and 2 cost nothing. Pick one step, do it for two weeks, and re-score in 90 days. The score describes recent habits, not your ability.
Does a high score mean I am ready for an AI role?
No. The score measures your habits in using AI at work. It says nothing about whether any employer will hire you or pay you more, and this guide makes no promise of that kind.
Is this the same as the Garage Labs AI readiness quiz?
No. The quiz at /ai-quiz is a separate tool with its own questions and scoring. The 12 questions in this guide are an independent rubric, and the scores do not convert between them.
How long does the assessment take?
About 15 minutes if you score from memory, longer if you look up evidence for every 4 and 5. The evidence step is the part that makes the score useful, so it is worth the extra time.
Where to go from here
To see new guides as they publish, join the Garage Labs community. If you want a quick outside view, the separate AI readiness quiz is there too.
If you have 5 or more years of professional experience and want to move from using AI to building with it, the Applied AI Practitioner Program has you build an AI product copilot, a RAG (Retrieval-Augmented Generation) knowledge assistant with citations, an autonomous research agent and an evaluation harness, plus a capstone, through live sessions and build labs. It is designed for experienced professionals such as product managers, product leads, engineers, analysts and operators. It is not for students or people without professional experience, anyone looking for a job guarantee, or anyone looking for a tool tour. If that is not you yet, the free steps for your level above are a better use of the next 90 days. This assessment is not an entry test for the programme. See the programme page for the current schedule and details.
Found this useful? Share it with someone who needs it. Visit garagelabstech.com and follow Sumit Kumar Singh for more on LinkedIn.
Sources and further reading
- US District Court, Southern District of New York, Mata v. Avianca, Inc., No. 22-cv-1461 (PKC), Opinion and Order on Sanctions (22 June 2023; copy hosted by the US District Court for the District of New Hampshire)
- NIST (National Institute of Standards and Technology), Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile, NIST AI 600-1 (July 2024)
- Cisco, Cisco AI Readiness Index, 2023 edition report (an example of an organisation-level assessment)
- DLA Piper, Data protection laws of the world: India
- Anthropic, AI Fluency: Framework & Foundations (free course from an AI developer)
- Your employer's AI policy and the current data terms of the tools you use. Terms change by provider, plan and date, and nothing in this guide quotes them.