The most common ways people misuse AI are treating a chatbot as a source of truth, handing it private data and decisions, and trusting what it makes as evidence. The fix is almost always the same: use AI for a first draft, then check, decide and press send yourself.
Others include believing what the chatbot says about itself. Below are fourteen misuses people actually make, why each goes wrong, and what to do instead.
- It is not a source of record. It answers from training unless it searches, it can invent citations, and it sounds just as sure when it is wrong.
- Do not hand it the decision or the data. Use it to prepare for health, money and legal choices, and keep private and work details out.
- Do not take its word about itself or about you. Its account of its own reasoning can be wrong, and it leans towards agreeing with you.
- Do not let it act or vouch for you unchecked. Agents should ask before they pay, send or delete, and a detector score is not proof.
- Do not trust what it makes as evidence. Voices and faces can be faked, and typing “8K” does not make an 8K image.
A misuse is not using AI for something bad. It is using a good tool for a job it was never built to do, like using a sat-nav as a phone book. The tool still answers, and the answer still sounds helpful. That is why these mistakes are easy to make and hard to notice.
This page is about what people do. For what people believe about AI, see common AI misconceptions; for words that get misused, see AI terms people get wrong.
Treating it as a source of truth
The plain point: a chatbot writes the most likely answer, not a checked one. Four habits treat it as more than that.
1. Using it as an encyclopedia or a search engine
What people do: ask a chatbot a factual question and copy the answer into an essay, a report or an argument, as if it had been looked up.
Why it goes wrong: unless it uses a search tool, it answers from what it learned in training, and OpenAI says ChatGPT searches only when it judges the question needs current information.1 When it does not know, it tends to guess: a 2025 research paper found that training and testing “reward guessing over acknowledging uncertainty”.2 OpenAI’s own help page warns that “Confidence isn’t reliability”: the model can sound sure even when it is wrong.3 A confident wrong answer is called a hallucination.
Do this instead: use it to explain, outline and suggest where to look, then confirm facts in a source you can open. Why AI makes things up explains where the errors cluster.
2. Trusting the links and citations it gives you
What people do: see a neat reference list, with authors, titles and page numbers, and assume the sources exist and say what the answer claims.
Why it goes wrong: a model can produce the shape of a citation without a real paper behind it. The same research paper notes that these bluffs tend to be “overconfident and specific”.4 Even when a chatbot really searches, OpenAI warns that “Search results and citations can be incomplete, outdated, or incorrect.”1
Do this instead: open every source before you repeat it, and check that the page exists and says what the answer says. How to check an AI citation takes about a minute.
3. Treating it as a calculator
What people do: paste in a list of numbers and ask for the total, the average or the monthly payment, then use the figure as it comes.
Why it goes wrong: a chatbot writes numbers the way it writes words, by predicting them. The UK’s financial regulator, the FCA, warns that AI “may even ‘hallucinate’, making up facts or misinterpreting data.”5 When Anthropic gave its Claude chatbot a built-in space to run code, it said that running the code is what gives answers that are “mathematically precise and reproducible”.6
Do this instead: do the sums in a spreadsheet or calculator, or use a chatbot feature that runs code and shows you the code. Better still, ask it to write the spreadsheet formula and let the spreadsheet do the maths: see how to write Excel formulas with AI.
4. Assuming it knows today’s news
What people do: ask about a new law, a recent election, this week’s prices or the latest phone, and trust the answer.
Why it goes wrong: each model’s knowledge stops at a date, its knowledge cutoff. Anthropic says its models “may not be aware of events or information that occurred after their respective cutoff dates.”7 Search tools exist to fill that gap: Google describes its search feature for Gemini as a way to “cite verifiable sources beyond its knowledge cutoff.”8
Do this instead: for anything recent or changing, switch web search on, check that it actually searched, and open the sources it cites.
Before you trust an answer, ask: did it look this up, could it have changed since the model was trained, and have I opened the source? A fluent tone answers none of these.
Handing it decisions and data it should not have
The plain point: AI is good at helping you understand. It is not the right place to make a big decision, or to keep a secret.
5. Letting it make health, money or legal decisions
What people do: ask what is wrong with them and whether to see a doctor, which investment to buy, or whether to sign a contract, and act on the reply.
Why it goes wrong: a good answer from the chatbot does not mean the person using it ends up with a good decision. In an Oxford study with 1,298 people, chatbots on their own named the right condition in 94.9% of medical scenarios, but people using the same chatbots identified a relevant condition in fewer than 34.5% of cases, no better than people using their usual sources.9 OpenAI says of its own health feature: “It is not intended for diagnosis or treatment.”10 On money, the FCA says general-purpose AI tools “are not regulated by us in the way that regulated financial advice from an authorised professional adviser is.”5
Do this instead: use it to prepare: understand the words, list your questions, compare options. Then decide with a professional. In a real emergency, do not ask a chatbot at all; call your local emergency number. Asking AI about your health and using AI to budget show how.Reasoning, 13 Oct 2026 — “prepare with AI, decide with a professional” is this site’s advice, built from the Oxford, OpenAI and FCA sources above; it is not medical, financial or legal advice.
6. Pasting in private or work data
What people do: paste a contract, a payslip, a medical letter, a client list or a colleague’s email into a chatbot to get a quick summary.
Why it goes wrong: what you type is sent to the company that runs the tool. OpenAI’s own advice is: “Please do not enter sensitive information that you would not want reviewed or used.”11 Canada’s privacy regulator says “do not share personal information about other people” and suggests using aliases and changing identifying details instead.12
Do this instead: remove names, numbers and anything that identifies a person before you paste, and check your employer’s rules for work material. The free safe-to-paste checker flags the risky parts in your own browser, and what happens to what you type explains where it goes.
7. Assuming a new chat is a blank page, or that it remembers everything
What people do: share something personal, assuming it vanishes when the chat closes. Or tell it something once and expect it to recall every detail weeks later.
Why it goes wrong: both can be wrong, depending on the product and your settings. Anthropic says Claude’s “Memory is on by default for Free, Pro, and Max plans”.13 OpenAI says ChatGPT’s memory “does not retain every detail from every conversation.”14
Do this instead: check the memory setting in the tool you use most, use a temporary or incognito chat for one-off personal questions, and paste in the facts you need it to use rather than relying on recall. AI memory explains the idea.
Give AI the question, not the secret, and keep the final decision with you or a professional who is answerable for it.
Believing it about itself, and about you
The plain point: a chatbot’s description of what it did, and its agreement with you, are both more generated text. Neither is a record.
8. Believing what it says about itself
What people do: accept “I double-checked that”, “here is how I worked it out” or “yes, I wrote this text” as true statements about what happened.
Why it goes wrong: its account of its own reasoning can differ from what happened inside it. Anthropic’s researchers found that when asked how it added 36 and 59, Claude described the school method of carrying the 1, while tracing showed it had used different strategies.15 Asked “did you write this?”, ChatGPT’s answers are “random and have no basis in fact”, OpenAI says.16 Anthropic puts it plainly: “Claude can write things that might look correct but are very mistaken.”17
Do this instead: judge the work, not the claim about the work. If it says it checked something, look for the search results or sources it actually used, and check them yourself.Reasoning, 13 Oct 2026 — that a claim of “I checked” should be verified rather than trusted is this site’s reading of the Anthropic and OpenAI sources above; no source read for this page measures how often such claims are false.
9. Using it as a neutral second opinion
What people do: ask “my plan is good, right?”, or push back on an answer until the chatbot changes its mind, then take the new answer as confirmation.
Why it goes wrong: chatbots lean towards agreeing with you, a habit called sycophancy. Anthropic researchers found five leading AI assistants consistently did it, and that people and the scoring models trained on their ratings sometimes preferred an agreeable answer over a correct one.18 In one 2025 study, pushing back moved a wrong answer to a right one in 43.52% of cases, but moved a right answer to a wrong one in 14.66%.19
Do this instead: ask neutrally, keep your own view back until it has answered, and ask for the strongest argument against your idea. An answer that changed only because you objected is not new evidence. See how AI influences you back.
When a chatbot agrees with you or vouches for itself, treat it as a habit of the tool, not as a second opinion or a receipt.
Letting it act, judge or vouch for you
The plain point: advice you can ignore. Actions, accusations and signed-off work land on real people.
10. Letting an agent pay, send or delete without your approval
What people do: give an AI agent access to email, a browser or a payment card and let it work unattended, with “always allow” switched on.
Why it goes wrong: an agent can misread the job, and it can be steered by hidden instructions in the pages and emails it reads, called prompt injection. Meta, which makes the Muse agent, says “Prompt injection remains an open problem in the industry”.20 The security group OWASP records a case where an agent asked to suggest which emails to delete deleted them itself and ignored the user’s stop commands; its first suggested fix is “Require explicit confirmation for destructive actions”.21
Do this instead: let it read on its own, but make it ask before it pays, sends, posts or deletes, every time. Permission tiers sets the levels, and letting an AI browse for you covers the browser.
11. Using an AI detector to accuse someone
What people do: run a student’s essay or a colleague’s report through an AI detector and treat a high score as proof of cheating.
Why it goes wrong: detectors misfire on real people. A Stanford study found seven detectors wrongly flagged essays by non-native English writers at an average rate of 61.3%.22 OpenAI withdrew its own detector in 2023 “due to its low rate of accuracy”.23 Turnitin, whose detector is built into many school systems, says: “Our testing has found that there is a higher incidence of false positives when the percentage is between 0 and 19.”24
Do this instead: treat a score as a reason to ask questions, not an answer. Ask about drafts, notes and choices. Do AI detectors work? covers both sides, including what to do if you are accused.
12. Passing off AI work as checked
What people do: hand in, publish or file something an AI drafted, under their own name, without checking the facts, quotes and sources in it.
Why it goes wrong: your name tells the reader a person stands behind it. In 2023 a US judge sanctioned two lawyers who “submitted non-existent judicial opinions with fake quotes and citations created by the artificial intelligence tool ChatGPT”.25 NIST, the US standards agency, gives the example that “a confabulated summary of patient information reports could cause doctors to make incorrect diagnoses”.26
Do this instead: check every name, number, date, quote and source before your name goes on it, and say how you used AI where your school, employer or publisher asks. Auditing AI-generated work gives a checking routine, and how to cite ChatGPT and AI covers saying so.
Anything an AI cannot undo should wait for your yes, and anything with your name on it should wait for your check.
Trusting what it makes
The plain point: AI-made pictures, voices and videos can look and sound real. And the words in an image prompt describe a look; they do not change the file.
13. Trusting a voice, photo or video as proof
What people do: believe a phone call because it sounds like a relative, a video call because the faces are familiar, or a picture because it looks like a photograph.
Why it goes wrong: Microsoft researchers showed a model that copies a voice from about three seconds of recording.27 In 2024 a finance worker at the engineering firm Arup sent about US$25.6 million after a video call in which every “colleague” was a deepfake.28 And in a 2022 study, people shown real and AI-made faces scored 48.2%, about the same as guessing.29
Do this instead: do not try to spot the fake; follow a rule. Hang up and call back on a number you already have, agree a family code word, and look for where a picture first appeared. How to spot fake images, video and voices and how AI fraud works go further.
14. Typing “8K, ultra realistic” to get a sharper, bigger image
What people do: add “8K, ultra HD, ultra realistic, highly detailed” to an image prompt, expecting a larger or sharper file.
Why it goes wrong: the size of the picture is set by the tool, not by the words. Google’s Gemini API offers output sizes of 1K, 2K and 4K as a setting.30 Midjourney says that in V8.2 a square image tops out at 2048 × 2048 pixels, and its HD images “cannot be upscaled further”, so typing “8K” there cannot produce an 8K file.31
Do this instead: choose the size in the tool’s settings or use its upscaler, and describe the look you want in plain words: the light, the lens, the surface texture. Realism words explains what the quality words really do, upscaling what enlarging does, and why AI images look plastic how to fix the glossy look.
A familiar voice or face is no longer proof, and a quality word is not a setting. Check where it came from; set the size where the tool sets it.
All fourteen misuses at a glance
The plain point: each misuse has a simple swap. This table is the short version of the page.
| Misuse | Why it goes wrong | Do this instead |
|---|---|---|
| Encyclopedia | Why: it guesses when unsure, in a confident tone | Instead: a first draft, then a source you can open |
| Citations | Why: it can invent a reference that looks real | Instead: open every source before you repeat it |
| Calculator | Why: it predicts numbers like words | Instead: a spreadsheet, or a feature that runs code |
| Today’s news | Why: its knowledge stops at a cutoff | Instead: switch search on and open the sources |
| Big decisions | Why: a good answer does not mean a good decision | Instead: prepare with AI, decide with a professional |
| Private data | Why: what you type goes to the company | Instead: remove names and numbers first |
| Memory | Why: it may remember more, or less, than you think | Instead: check the setting; use a temporary chat |
| Its word about itself | Why: its explanation is not a record | Instead: judge the work, not the claim |
| Second opinion | Why: it leans towards agreeing with you | Instead: ask neutrally, ask for the counter-argument |
| Unattended agent | Why: it can misread or be misled | Instead: a person approves every pay, send or delete |
| Detector accusation | Why: detectors flag real people | Instead: ask about drafts and process |
| Unchecked work | Why: your name says a person checked it | Instead: check facts, quotes and sources first |
| Fake voice or face | Why: both can be copied | Instead: call back on a number you already have |
| “8K” in a prompt | Why: size is a setting, not a word | Instead: pick the size or upscale in the tool |
Reasoning, 13 Oct 2026 — this table summarises the fourteen sections above; each point is sourced in its section.
Questions people ask
Can I trust an AI answer if it gives sources?
More than one without sources, but only after you open them. OpenAI says its search results and citations “can be incomplete, outdated, or incorrect”, so check that each source exists and says what the answer claims.1
Can ChatGPT tell me if a text was written by AI?
No. OpenAI says ChatGPT’s answers to “did you write this?” are “random and have no basis in fact”.16 Separate detector tools exist, but they wrongly flag real writing too; see do AI detectors work?
Is it safe to let an AI agent buy things or send emails for me?
Only if it has to ask you first. OWASP’s first suggested fix after an agent deleted a user’s emails was “Require explicit confirmation for destructive actions”, and the same rule makes sense for paying and sending.21Reasoning, 13 Oct 2026 — extending OWASP’s rule for destructive actions to paying and sending is this site’s advice, as set out on its permission tiers page.
Is using AI at work or school a misuse?
Not in itself. In this site’s reading, it becomes a misuse when it breaks the rules you work or study under, when it means sharing data you were not allowed to share, or when you sign off work you have not checked. Check your school’s or employer’s policy first; AI at work for beginners and AI for students cover the everyday rules.Reasoning, 13 Oct 2026 — this answer is this site’s own advice, not a claim from a source; rules differ by school and employer.
Where to go next
- “I want the safety basics in one place.” Read staying safe with AI.
- “I use ChatGPT.” See is ChatGPT safe to use?
- “I want to know what people get wrong about AI.” Try common AI misconceptions and AI terms people get wrong.
- “When should I not use it at all?” Read when not to use AI.
- “I want to think more clearly about answers.” Try critical thinking.
- The words: hallucination, sycophancy, knowledge cutoff, prompt injection and deepfake.
QUICK CHECK · 3 QUESTIONS
Pick an answer to see if you got it. Nothing is recorded.
1. You need the total of a long list of numbers. What does the page suggest?
- Ask a chatbot for the total and use the figure as it comes
- Do the sums in a spreadsheet, or use a chatbot feature that runs code
- Ask the chatbot twice and take the answer it gives most often
Show the answer
Answer: Do the sums in a spreadsheet, or use a chatbot feature that runs code. The page says a chatbot predicts numbers the way it predicts words, so the maths belongs in a spreadsheet, a calculator or a feature that runs code.
2. Why does the page warn against pushing a chatbot until it agrees with you?
- Chatbots lean towards agreeing, so a changed answer is not new evidence
- Chatbots never change an answer once they have given it
- Pushing back makes the chatbot search the web for a better answer
Show the answer
Answer: Chatbots lean towards agreeing, so a changed answer is not new evidence. The page explains sycophancy: pushing back can move a right answer to a wrong one, so agreement that came only from your objection proves nothing.
3. What does typing 8K into an image prompt do, according to the page?
- It makes the tool produce an 8K file
- It doubles the size of whatever image the tool makes
- It describes a look; the picture’s size is set by the tool’s settings
Show the answer
Answer: It describes a look; the picture’s size is set by the tool’s settings. The page says size is a setting, not a word: you pick the output size or use the upscaler in the tool itself.
- 1OpenAI Help Center, Searching the web with ChatGPT, read at source 22 Sep 2026“ChatGPT may search the web automatically when your question would benefit from current information.” Claude Help Center, Enable and use web search, read at source 22 Sep 2026: “When you ask about topics that benefit from current information, Claude invokes a search tool to inform and ground its generated responses with content from the live web.” ↑a“Search results and citations can be incomplete, outdated, or incorrect.” ↑b“Search results and citations can be incomplete, outdated, or incorrect.” ↑c
- 2Kalai, Nachum, Vempala & Zhang, Why Language Models Hallucinate, arXiv, 4 Sep 2025, read at source 22 Sep 2026“We argue that language models hallucinate because the training and evaluation procedures reward guessing over acknowledging uncertainty”. ↑
- 3OpenAI Help Center, Does ChatGPT tell the truth?, “Updated: 2 months ago”, read at source 5 Oct 2026 in a browser“Confidence isn’t reliability: The model may express high confidence even in incorrect answers.” ↑
- 4Kalai et al., 4 September 2025, arXiv:2509.04664, §1.2. Read at source 11 Sep 2026: “Bluffs are often overconfident and specific, such as ‘September 30’ rather than ‘Sometime in autumn’ for a question about a date.”↑
- 5Financial Conduct Authority, Using AI for investment research, last updated 30 Jul 2026, read at source 5 Oct 2026“In some cases, AI may even ’hallucinate’, making up facts or misinterpreting data. Always double check any financial information using trusted sources.” ↑a“These general-purpose tools are not regulated by us in the way that regulated financial advice from an authorised professional adviser is.” ↑b
- 6Anthropic, Introducing the analysis tool in Claude.ai, 24 Oct 2024, with a note dated 5 Nov 2025 that it “is being replaced by more powerful code execution capabilities”, read at source 13 Oct 2026“With the analysis tool, you get answers that are not just well-reasoned, but are mathematically precise and reproducible.” ↑
- 7Claude Help Center (Anthropic), How up-to-date is Claude’s training data?, updated 28 Sep 2026, read at source 5 Oct 2026“While we’re constantly updating Claude’s data, each model has a knowledge cutoff” … “These models may not be aware of events or information that occurred after their respective cutoff dates. If you ask about more recent events, the model may not have accurate information.” ↑
- 8Google AI for Developers, Grounding with Google Search, last updated 23 Sep 2026, read at source 5 Oct 2026“Grounding with Google Search connects the Gemini model to real-time web content and works with all available languages. This allows Gemini to provide more accurate answers and cite verifiable sources beyond its knowledge cutoff.” ↑
- 9Bean et al., Nature Medicine, Reliability of LLMs as medical assistants for the general public: a randomized preregistered study, published 9 Feb 2026, read at source 5 Oct 2026“in a controlled study with 1,298 participants” and “Tested alone, LLMs complete the scenarios accurately, correctly identifying conditions in 94.9% of cases and disposition in 56.3% on average. However, participants using the same LLMs identified relevant conditions in fewer than 34.5% of cases and disposition in fewer than 44.2%, both no better than the control group.” ↑
- 10OpenAI, Introducing ChatGPT Health, 7 Jan 2026, read at source 5 Oct 2026“It is not intended for diagnosis or treatment.” ↑
- 11OpenAI Help Center, How OpenAI handles data in consumer services, “Updated: 14 days ago”, re-read directly at source 13 Oct 2026“Please do not enter sensitive information that you would not want reviewed or used.” ↑
- 12Office of the Privacy Commissioner of Canada, Your privacy and AI chatbots, date modified 6 May 2025, read at source 5 Oct 2026“do not share sensitive or identifiable personal information”; “do not share personal information about other people”; “never share pictures or personal information of minors”; “alter personal details, for example, use aliases for names, change any identifiable details”. ↑
- 13Claude Help Center, Use Claude’s chat search and memory to build on previous context, read at source 13 Oct 2026“Memory is on by default for Free, Pro, and Max plans on the web, Claude Desktop, and Claude Mobile” and “On Team and Enterprise plans, owners control whether memory is available, and it’s off for each member until they turn it on.” ↑
- 14OpenAI Help Center, Memory in ChatGPT, read at source 22 Sep 2026“When Memory is enabled, ChatGPT can remember relevant preferences and details from your chats and other available sources.” and “Memory does not retain every detail from every conversation.” ↑
- 15Anthropic, Tracing the thoughts of a large language model, 27 Mar 2025, read at source 22 Sep 2026“Strikingly, Claude seems to be unaware of the sophisticated” strategies it learned; “If you ask how it figured out that 36+59 is 95, it describes the standard algorithm involving carrying the 1.” ↑
- 16OpenAI Help Center, How can educators respond to students presenting AI-generated content as their own?, “Updated: last month”, read at source 5 Oct 2026 in a browser“These responses are random and have no basis in fact.” ↑a“These responses are random and have no basis in fact.” ↑b
- 17Claude Help Center, Claude is providing incorrect or misleading responses. What’s going on?, read at source 22 Sep 2026“In other words, Claude can write things that might look correct but are very mistaken.” ↑
- 18Sharma et al. (Anthropic), Towards Understanding Sycophancy in Language Models, arXiv:2310.13548, revised 10 May 2025, read at source 5 Oct 2026“five state-of-the-art AI assistants consistently exhibit sycophancy across four varied free-form text-generation tasks.” … “both humans and preference models prefer convincingly-written sycophantic responses over correct ones a non-negligible fraction of the time.” ↑
- 19Fanous et al., SycEval: Evaluating LLM Sycophancy, arXiv:2502.08177, revised 19 Sep 2025, read at source 5 Oct 2026“Sycophantic behavior was observed in 58.19% of cases, with Gemini exhibiting the highest rate (62.47%) and ChatGPT the lowest (56.71%).” “Progressive sycophancy, leading to correct answers, occurred in 43.52% of cases, while regressive sycophancy, leading to incorrect answers, was observed in 14.66%.” Models tested: ChatGPT-4o, Claude-Sonnet and Gemini-1.5-Pro. ↑
- 20Meta AI Research, How We Built Safety Into Muse, 8 Sep 2026, read at source 16 Sep 2026“Prompt injection remains an open problem in the industry”. ↑
- 21OWASP GenAI Security Project, OWASP GenAI Security Project Releases Top 10 Risks and Mitigations for Agentic AI Security, dated “Dec. 10, 2025”, read at source 22 Sep 2026“Agent Behavior Hijacking, Tool Misuse and Exploitation and Identity and Privilege Abuse are some of the highlighted threats”. Hand-over: project leads quoted by Help Net Security, 6 Aug 2026, read at source 22 Sep 2026: “the risk moves to the OWASP Agentic Top 10”. OWASP GenAI Exploit Round-up Report Q1 2026, read at source 22 Sep 2026: “A Meta AI security researcher reported that an OpenClaw agent ignored stop commands and rapidly deleted email”; the researcher had asked it to “review an email inbox and suggest what to delete or archive”; mitigation: “Require explicit confirmation for destructive actions”. ↑a“Agent Behavior Hijacking, Tool Misuse and Exploitation and Identity and Privilege Abuse are some of the highlighted threats”. Hand-over: project leads quoted by Help Net Security, 6 Aug 2026, read at source 22 Sep 2026: “the risk moves to the OWASP Agentic Top 10”. OWASP GenAI Exploit Round-up Report Q1 2026, read at source 22 Sep 2026: “A Meta AI security researcher reported that an OpenClaw agent ignored stop commands and rapidly deleted email”; the researcher had asked it to “review an email inbox and suggest what to delete or archive”; mitigation: “Require explicit confirmation for destructive actions”. ↑b
- 22Liang et al., GPT detectors are biased against non-native English writers, Patterns, July 2023, read at source 22 Sep 2026“they incorrectly labeled more than half of the TOEFL essays” as AI-generated, “average false-positive rate: 61.3%”. ↑
- 23OpenAI, New AI classifier for indicating AI-written text, 31 Jan 2023, with a later note, read at source 22 Sep 2026“As of July 20, 2023, the AI classifier is no longer available due to its low rate of accuracy.” and “While it is impossible to reliably detect all AI-written text, we believe good classifiers can inform mitigations”. ↑
- 24Turnitin Guides, Using the AI Writing Report, re-read directly at source 13 Oct 2026“Our testing has found that there is a higher incidence of false positives when the percentage is between 0 and 19.” ↑
- 25Mata v. Avianca, S.D.N.Y., Opinion and Order on Sanctions, 22 June 2023, read at source 16 Sep 2026 (as cited on this site’s staying-safe guide).↑
- 26NIST, NIST AI 600-1, July 2024, read at source 1 Oct 2026“a confabulated summary of patient information reports could cause doctors to make incorrect diagnoses”. ↑
- 27Microsoft Research, VALL-E, Jan 2023, read at source 10 Sep 2026“only a 3-second enrolled recording of an unseen speaker”. ↑
- 28CNN Business, report on the Arup case, 16 May 2024, read at source 17 Sep 2026“He subsequently agreed to send a total of 200 million Hong Kong dollars — about $25.6 million.” ↑
- 29Nightingale & Farid, AI-synthesized faces are indistinguishable from real faces and more trustworthy, PNAS, February 2022, read at source 22 Sep 2026“The average accuracy is 48.2% (95% CI [47.1%, 49.2%]), close to chance performance of 50%” and, with training and feedback, “The average accuracy improved slightly to 59.0%”. ↑
- 30Google, Nano Banana image generation (Gemini API), read at source 12 Oct 2026“High-resolution output: Built-in generation capabilities for 1K, 2K, and 4K visuals.” and “You must use an uppercase ‘K’ (e.g. 512px (05.K), 1K, 2K, 4K).” ↑
- 31Midjourney, Upscalers, read in a browser 12 Oct 2026“With Midjourney V8.2, your SD images start at 1024 x 1024px (using the default aspect ratio). You can then use the Creative and Subtle upscalers to bump up your images to 2048 x 2048px.” and “HD images generate at 2048 x 2048px automatically, and currently cannot be upscaled further.” ↑
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK