While Big Tech continues pouring billions into AI infrastructure, a BBC investigation revealed that nearly half of all AI responses contain errors. Meanwhile, deepfake fraud has quadrupled in six months, netting criminals close to a billion dollars. This is the reality check nobody wanted.
The BBC Dropped a Truth Bomb: AI Is Wrong 45% of the Time
The European Broadcasting Union published research that should terrify anyone relying on AI for information: 45% of news queries to ChatGPT, Microsoft Copilot, Gemini, and Perplexity produce erroneous answers. The systems incorrectly identified the Pope, got the German Chancellor wrong, and when asked about bird flu concerns, Copilot confidently cited an Oxford vaccine trial—sourcing a BBC article from 2006, nearly 20 years old.
Perplexity claimed surrogacy is “prohibited by law” in Czechia when it’s actually unregulated—neither explicitly banned nor permitted. Gemini mischaracterized vape legislation, claiming buying would be illegal when only sales and supply were being banned. AI researcher Josh Bersin ran an experiment asking ChatGPT to analyze AI data center capital investments and got results claiming there are more AI engineers in the US than total working people.
The problem is structural: even if only 2% of input data contains errors, many queries will produce poor results. And as OpenAI and Google push toward advertising business models, these systems are becoming less trustworthy, not more. “Unless you’re using a highly trusted corpus, you as a user must verify the answers yourself,” Bersin warns.
Deepfake Fraud Hit $897 Million in Six Months
The first half of 2025 saw 580 deepfake incidents—nearly four times the entire 2024 total of 150. With just 15 seconds of audio, scammers can now clone voices using tools like ElevenLabs, Speechify, and Resemble AI. Criminals are using generative AI tools like DeepFaceLive, Magicam, and Amigo AI to alter their face, voice, gender, and race during live video calls.
https://www.scamwatchhq.com/deepfake-deception-the-897-million-ai-scam-revolution-threatening-everyone-in-2025/
More than 50% of fraud now involves artificial intelligence, and 1 in 20 identity verification failures are linked to deepfake attacks. In the US, deepfake-related identity fraud jumped from 0.2% to 2.6% between 2022 and early 2023—and it’s only accelerated in 2025.
The most spectacular case: British engineering firm Arup lost $25 million after a video conference where deepfaked versions of the CFO and other employees convinced a real employee to make 15 transfers totaling $26 million. This wasn’t pre-rendered synthetic video—these were real-time deepfakes improvising and adapting on the fly to bypass biometric checks.
Lawyers Filing Court Documents With AI Hallucinations
A major US law firm is “profoundly embarrassed” after one of its attorneys submitted a court filing riddled with fake citations and inaccuracies generated by AI. The firm apologized to the judge and accepted any imposed sanctions. They’ve since updated their AI policies—but the damage is done. When legal professionals can’t distinguish between real case law and AI fabrications, we’ve got a systemic problem.
Mistake-filled legal briefs show the limits of relying on AI tools at work
The ‘Homeless Man Prank’ Won’t Stop
US police departments continue fielding 911 calls triggered by an AI prank: teenagers generate images of disheveled “homeless men” and send them to parents claiming the person entered their home. Departments in Michigan, New York, Wisconsin, and Massachusetts have issued official warnings calling the trend “stupid and potentially dangerous”.
https://www.forbes.com/sites/lesliekatz/2025/10/17/viral-ai-homeless-man-prank-condemned-by-police-and-advocacy-groups/
The Yonkers Police Department in New York explained the problem: “Officers are responding FAST using lights-and-sirens to what sounds like a real intruder—and only getting called off once everyone realizes it was a joke. That’s not just a waste of resources… it’s a real safety risk for officers and families if we get there before the prank is revealed and rush into the home to apprehend this ‘intruder’ that doesn’t exist”.
https://abc11.com/post/ai-homeless-man-prank-police-issue-new-warning-trend-faking-intruder-home/18018899/
Cathy Alderman from the Colorado Coalition for the Homeless added crucial context: “Individuals experiencing homelessness are significantly more likely to be victims of violent crimes due to their vulnerability, rather than being violent offenders themselves. This video trend is likely to exacerbate the situation by suggesting that those experiencing homelessness are a threat”.
Nearly Every Company Is Losing Money on AI
An EY survey found that nearly every large company that introduced AI has incurred initial financial losses. The culprits: compliance failures, flawed outputs, algorithmic bias, and disruptions to sustainability goals. The Bank of England warned that while data centers are currently funded by the largest companies, there will likely be growing reliance on debt—and if AI fails to deliver or requires less computing power than anticipated, risks could escalate.
https://www.nytimes.com/2025/10/31/technology/ai-spending-accelerating.html
China Wants to Control Global AI Governance
At the APEC summit, Chinese President Xi Jinping proposed creating a World Artificial Intelligence Cooperation Organization to set governance rules and position AI as a “public good”. This is a direct challenge to US technological leadership: China is positioning itself as an alternative on trade cooperation and technology development. APEC adopted declarations on AI and demographic change—but Xi got center stage to push his geopolitical agenda.
https://www.reuters.com/world/china/chinas-xi-pushes-global-ai-body-apec-counter-us-2025-11-01/
What This Actually Means
AI is wrong 45% of the time, but the industry keeps spending billions. Deepfake scammers netted nearly a billion dollars in six months. Lawyers are filing court documents with fabricated citations. Teenagers are pranking parents and diverting police resources. Companies are losing money deploying AI. And China wants to create a global organization to control the rulebook.
We’re living in a world where half of AI answers are wrong and the other half might be fraud. And everyone’s pretending this is fine.
No comment