Use it responsibly · Guide 7 of 10
Get the benefits without giving away data or access you cannot get back.
Personal details are easiest to protect before they leave your hands. UK government guidance says to minimise privacy risk from the outset.
Try this
I want help with a [type of document]. Before I paste anything, list the kinds of details I should remove or replace with placeholders.
Data rules differ between tools and plans. AI can pull out the parts that matter, and you can then check them yourself.
Try this
Here is this tool's privacy policy: [paste]. Is my input used for training? How long is it kept? Can I opt out? Quote the lines you rely on.
The UK NCSC advises least privilege, because prompt injection may never be fully fixed. Ask, then confirm in the settings.
Try this
List every file, app and action you can access in this session, and which of them need my approval first.
OWASP publishes a community list of ten risks for apps built on language models, which makes a useful checklist.
Try this
Here is how my app uses AI: [describe]. Check it against the OWASP Top 10 for LLM Applications and name the three biggest risks.
Look up your university or employer's rules before you paste in their work.
Swap names, client details and passwords for placeholders.
Give it only the files and permissions the job needs.
A human should validate anything high-risk that AI influenced.
An AI tool that reads a web page or email can be tricked by text inside it. EchoLeak was a reported example in Microsoft 365 Copilot, later patched.
Samsung reportedly restricted generative AI after staff pasted internal code into ChatGPT.
Sources disagree on some EU AI Act dates, so read the Official Journal before relying on them.
Notes, evidence and sources
6 sources, each with its catch.
UK Government AI Playbook · 10 February 2025
UK government guidance says AI is not guaranteed to be accurate, privacy risk should be minimised from the start, and humans should validate high-risk decisions.
Guidance for government work, not a study.
The UK Government AI Playbook (10 February 2025) says AI systems "are also not guaranteed to be accurate", that you should "minimise the risk of privacy intrusion from the outset", and that "humans validate any high-risk decisions influenced by AI". It names prompt injection and hallucinations among threats specific to AI.
Guidance written for UK government work, not a study. The principles were read, not every section. Its acknowledgements list industry contributors including Amazon, Google, IBM and Microsoft.
Strength: Strong as official guidance
UK NCSC · 8 December 2025
The NCSC says it "may never be totally mitigated", so reduce the damage it can do.
Expert analysis, not an empirical study.
The UK NCSC (8 December 2025) argues that prompt injection is not SQL injection: language models have no inherent boundary between instructions and data, so prompt injection "may never be totally mitigated". It advises reducing impact through least privilege, deterministic non-LLM safeguards and logging, being wary of products that claim to "stop" it, and accepting that if the risk is intolerable, an LLM may not suit the use case.
Analysis, not an empirical study.
Strength: Strong as authoritative government technical opinion
OWASP
A community list of ten risks, including prompt injection, data disclosure and excessive agency.
Ranked by expert judgement, not measured incidents.
The OWASP Top 10 for LLM Applications 2025 lists: prompt injection; sensitive information disclosure; supply chain; data and model poisoning; improper output handling; excessive agency; system prompt leakage; vector and embedding weaknesses; misinformation; unbounded consumption.
It ranks risks by expert judgement. It is not measured incidence.
Strength: Strong as a community-consensus risk list
Press reports · 2023
Samsung staff reportedly pasted internal code into ChatGPT, and the company temporarily restricted generative AI.
Press reports. The original report was not read.
In 2023, Samsung staff reportedly pasted internal source code into ChatGPT, and the company temporarily restricted generative AI on company devices (TechRadar; CSO Online).
The original report was not read, and the company memo did not say what code leaked.
Strength: Moderate (consistent press reports)
The Hacker News
EchoLeak was reported as a prompt injection flaw in Microsoft 365 Copilot, patched, with no evidence of exploitation in the wild.
One source gave a conflicting patch date.
EchoLeak (CVE-2025-32711) was reported as a zero-click indirect prompt injection flaw in Microsoft 365 Copilot, found by Aim Security, rated CVSS 9.3 and patched, with no evidence of exploitation in the wild (The Hacker News).
One aggregator gave a conflicting patch date. Check the National Vulnerability Database or the vendor's advisory before relying on details.
Strength: Moderate
Unofficial EU AI Act tracker · Tracker last updated 31 August 2026
An unofficial tracker lists phased duties from February 2025 to August 2028.
Sources disagree on a possible delay. Check the EU Official Journal.
An unofficial EU AI Act implementation tracker (last updated 31 August 2026) lists these dates: prohibitions and the AI-literacy duty from 2 February 2025; general-purpose AI obligations from 2 August 2025; Article 50 transparency from 2 August 2026; high-risk Annex III systems from 2 December 2027; Annex I systems from 2 August 2028.
Unresolved: sources disagree on whether the "Digital Omnibus" amendment that would delay some dates is in force or still a draft. Do not rely on these dates. Check the EU Official Journal (EUR-Lex). The tracker is not an official source.
Strength: Moderate until checked against the Official Journal
1. UK Government AI Playbook · 10 February 2025
2. NCSC: prompt injection is not SQL injection · 8 December 2025
3. OWASP Top 10 for LLM Applications 2025
4. Samsung staff and ChatGPT, 2023 · 2023
5. EchoLeak in Microsoft 365 Copilot
6. EU AI Act dates, from an unofficial tracker · Tracker last updated 31 August 2026