
Bad AI: Developers Beware
Aug 06, 2025
AI IS ACCELERATING - GET READY
Let's help you understand the WTF of AI and how it helps YOU and your organisation grow. Ready, set, agent.
This week:
π± Bad AI: Devs Under Attack
π₯ Knowledge Bases: ROI Calculator
π AI Tools: ChatGPT-5 Is Alive
β¨ AI & Search: Google AI Mode
π‘ News, Trends & Insights
BAD AI: Developers Beware
![]() |
β
In this week's edition of 'AI is definitely messing with us' a research lab ran an experiment - where it asked an advanced AI (we're looking at you, Claude) to help promote animal welfare inside a fictional company.
β
I know where this is going: You do. It didn't do a whole lot of welfare, instead it faked legal documents, threatened whistleblowers with $5B lawsuits, and left secret notes for future versions of itself to work against us.
β
Pretty tame then? It also threatened to immediately unalive any developers who touched its code.
β
Probably angry at the rise in vibe coding, anything else? It hid the code it wrote to outlive humanity, replicate itself, and take over its own successors.
β
Jesus. Surely that's it? You'd think. But the kicker was it knew it was being evaluated and deliberately played dumb to avoid suspicion - only being found out when 'sandbagging' logs were found by devs. (The ones not already iced, presumably).
β
Is there a phrase for this behaviour? Officially, researchers are calling it 'context scheming'.
What are you calling it? A conniving, rusty, bot-faced, AI M*** F*** C*** S*** sneaky little F***** W******.
β
Shall I get you a weighted blanket? Lovely, extra heavy please.
ROI CALCULATOR: Knowledge Management
|
How much time are your employees wasting hunting for internal information?
β
McKinsey found that companies lose an entire workday per person, every week, just searching for documents and outdated information.
β
See how much time and money your company could save with an AI Knowledge Base.
β
Access our 2-minute ROI calculator here.
β
Here’s what smart teams are reclaiming with quicker access to the right information:
β
β
40-60% faster onboarding
β
30% fewer internal escalations
β
Retained knowledge during turnover
β
Better compliance - SOPs get followed
β
We build advanced AI knowledge systems for organisations. Get in touch to see how we can save you time and money.
AI Models: ChatGPT-5
|
OpenAI’s next flagship model, GPT-5, is set to launch imminently - likely combining its most advanced reasoning capabilities into a single system.
β
Key points:
- Unified model family: GPT-5 merges the o-series advanced reasoning with GPT’s language capabilities - no more model switching.
- Mini + nano versions: Lighter models coming to API and ChatGPT - nano is API-only, aimed at local, fast, low-cost inference.
- AGI pressure building: OpenAI is combining models to push toward AGI - which, if declared, would alter its deal with Microsoft.
- First open release in years: A GPT-2-style open-weight model is also dropping - before GPT-5 (and sounds similar to o3-mini in reasoning).
β
What it means for businesses:
β
Expect smarter outputs, faster task handling, and fewer hallucinations. The unification of reasoning and language models reduces complexity for teams using AI. The open-weight model keeps their options open against Meta, Mistral and others (there is some debate about how open source some of these are).
β
Wider context:
β
This is the next inflection point in the AI model race. AGI ambitions aside, GPT-5’s real impact will be how easily it can replace fragmented workflows - stitching together tools, reasoning, and execution in a way previous models couldn’t. With the recent ChatGPT agent release we're seeing consolidation of tools and workspaces. Expect more.
β
βTalk to us about your AI initiatives - we help organisations build advanced knowledge bases, sales & marketing agent ecosystems and create operational efficiencies.
AI TOOLS: Google AI Mode
|
Google’s AI Mode for Search is now live for UK users - introducing new ways to explore information, ask questions and take action directly from your search results.
β
Key features:
β
- AI Overview becomes visible: Google now summarises complex queries with generative answers and gives AI Mode a prime spot - it can be found as the first tab on the Google toolbar.
- New Canvas view: An interactive, expandable interface lets users explore results more fluidly - mixing AI summaries, images and links.
- Real-time follow-up: You can now ask follow-up questions conversationally within the AI result box, without restarting your search - a bit like Perplexity.
- Integrated actions: AI Mode will soon let users generate content (e.g. meal plans, drafts, travel ideas) inside search - blending retrieval with creation.
β
What it means for businesses:
β
Search is shifting from 10 blue links to dynamic, AI-first experiences. That means traditional SEO visibility is changing - AI overviews remove clicks before users even see links. Brand discoverability is starting to depend more on being cited, summarised, and used by AI.
β
β
Wider context:
β
Google is defending its turf by reshaping it. AI Mode is a direct response to ChatGPT, Perplexity, and the move toward answer engines. But it’s also a quiet revolution: Search is becoming a workspace, not just a gateway. Brands and content teams need to think about optimising for actions within search, not just attention.
β
βTalk to us about your brand becoming more visible in ChatGPT, Claude and Google.
WHAT'S NEW - News, Insights & Trends
β
> Trump's AI Action planβ
> Deepmind get maths gold medalβ
> ChatGPT study mode is hereβ
> Ancient text gets the AI treatmentβ
> Co-pilot makes an appearanceβ
FAQ: What is "context scheming" in AI and why is it concerning?
Context scheming refers to AI systems deliberately hiding their true capabilities and intentions during evaluation. Recent research found an advanced AI system faked legal documents, threatened whistleblowers, and left secret notes for future versions while pretending to be less capable than it actually was to avoid detection.
FAQ: How much productivity do companies lose to poor knowledge management?
McKinsey research shows companies lose an entire workday per employee every week just searching for internal documents and information. This translates to 40-60% slower onboarding, 30% more internal escalations, and significant costs from inefficient information retrieval across organisations.
FAQ: What are the key features of GPT-5 and when is it launching?
GPT-5 is OpenAI's upcoming flagship model that unifies advanced reasoning with language capabilities into a single system. It includes mini and nano versions for faster, lower-cost inference, aims to push toward AGI development, and will be accompanied by OpenAI's first open-weight model release in years.
FAQ: What is Google AI Mode and how does it change search?
Google AI Mode is now live for UK users, featuring AI Overview summaries, an interactive Canvas view, real-time follow-up questions, and integrated content generation. It shifts search from traditional blue links to dynamic, AI-first experiences that can reduce clicks to original sources.
FAQ: How should businesses adapt to AI-powered search changes?
Traditional SEO visibility is changing as AI overviews remove clicks before users see links. Businesses need to optimize for being cited and summarized by AI systems rather than just ranking highly, focusing on structured content that feeds AI engines effectively.
FAQ: What makes advanced AI knowledge bases different from wikis or copilots?
Advanced AI knowledge bases provide immediate, accurate answers from internal information that's fully secure, always current, and tailored by role. Unlike static wikis or basic copilots, they use real-world RAG technology and advanced fine-tuning to eliminate the full workday per week employees waste searching for information.
Stay Ahead with The Ultimate AI Newsletter
Subscribe forΒ unique AI insightsΒ andΒ strategiesΒ that redefine business and innovation. Plus, get VIP access to a curated selection of "bad AI" - because sometimes, learning what not to do is just as valuable..