Pitfalls
Pitfalls for GenAI users
Common pitfalls
How to minimize risks
Pitfalls for GenAI users
Common pitfalls
How to minimize risks
Overview written by Martin Ågren - September 27, 2026
Training data = Prompts → training data? - Recommendation: Paid plan needed? + Check: Turned off - Data controls > "improve model..."
Knowledge cutoff = See infographic - LLM "frozen in time" - Recommendation: Event occurred after knowledge c? + Web search
Hallucination = AI outputs incorrect info - Reason: Reason: Insufficient training or no access to necessary information
- Reduce risk, most relevant: Web search / File upload / Reference text / One theme per chat: Avoid confusing info / Open-ended question
- Reduce risk, somewhat relevant: Short chat (short inputs & outputs) / English / Temperature, low: Favor high-probability output
- Identify: Your expertise: Easier to spot errors / "Search online again" / Links: Hover, is page relevant, check info / Ask AI to quote
Backup / Data loss, chats = Bugs in chatbots can make chats disappear - Recommendation: For important chat history: Backup to cloud
Write permission / Data loss, HDD = Permission to WRITE to your HDD? - Recommendation: Only grant write access when necessary
Task switching = Example: Switching tasks when waiting for autonomous agents to finish - Result: mental fatigue & reduced productivity
Cognitive debt = Cognitive skills lost (too much AI) - But is it really a problem? - Recommendation: Do tasks without AI regularly
Bias in AI = Consistent distortions in AI output - Recommendation: "What country?" + Compare LLMs + "Other perspectives"
Agreeable LLM = LLM is often agreeable - Ideas need to be challenged? - Recommendation: Neutral prompt + "Challenge assumptions"
Slop / AI slop = Low-quality, generic output - Recommendation: Take responsibility for your work: Check output, sources, quality
GDPR compliance = Personal data sent to server outside EU? - Recommendation: DPA agreement via Business/Ent + Local inf
Environmental impact = AI at scale uses a lot of electricity - Recommendation: Consider using small & fast model (or no AI)
Sovereignty = Over-reliant on US tech cloud services? - Recommendation: Consider: Hybrid concept with local inf / Mistral Vibe
Financial loss = AI agent with payment access? - Recommendation: Consider no payment access + Set spending limits + Require approval
Costs = Subscription & API calls (You can use ) - Recommendation: Spending limit for API + Monitor API usage + Avoid overkill
Timeline, knowledge cutoff