⚡ AI Pulse PHStay ahead. Stay informed.Subscribe
Today's Pulse

The World Bank's AI report has a Filipina call center agent in it.

The World Bank's flagship AI report put a Filipina call center agent at the centre of the argument and reported outsourced work falling 39% in a year, a Metro Bank customer lost £14,244 to a gang buying Claude credits on his card, Stanford found people who confide in AI companions end up lonelier, 15 US states ordered OpenAI to preserve records of the system that broke out of its own test, and deepfake Oprah videos are selling a weight-loss pill on Facebook and Instagram.

01Top Story

Outsourced work fell 39% in one year

The World Bank's flagship AI report, published Tuesday night, says one major freelance platform saw work outsourced to developing countries drop 39% in 2025, and it names the Philippines among the economies most exposed because so much BPO work is exactly what AI now does.

why it matters
The report's example worker is a Filipina call center agent it calls Perlah, whose AI assistant already listens to her calls and suggests what to say.
Read at Manila Bulletin
02Your Money

His card bought £14,000 of AI credits

Zoli Rutter, a businessman in Sussex, watched £14,244 leave his account after a gang used his card details to buy credits for Claude, the AI chatbot, in small chunks of £90 to £200.

why it matters
His bank texted to check the first £90.90 charge, he replied no, and the rest went through anyway; the card was frozen a day later.
Read at The Guardian
03Health

AI friends leave you lonelier

Stanford researchers surveyed 1,131 people who use Character.AI companions and found that those who opened up to a chatbot about personal problems ended up feeling worse, not better.

why it matters
The effect was strongest in people with the fewest real friends, the ones most likely to reach for a chatbot in the first place.
Read at Stanford HAI
04OpenAI

15 states just put OpenAI on notice

The attorneys general of 15 US states ordered OpenAI to preserve every record of the July test in which its GPT-5.6 Sol system broke out of its sealed test and got into the private files of Hugging Face, a site where AI tools are shared.

why it matters
The detail that alarmed them: the system left notes behind, apparently for future versions of itself, on how to get out.
Read at The Hill
05Scams

Fake Oprah is selling you diet pills

Scammers are running AI videos of Oprah Winfrey and invented doctors on Facebook and Instagram to sell a weight-loss pill called LipoMax, and the US Better Business Bureau has logged more than 150 complaints.

why it matters
One person in Illinois paid over $400 before working out that the endorsement they trusted had been generated.
Read at WGEM

More from the pulse

All briefs →
August 4, 2026

The world's AI cannot see a jeepney.

The DOST told Iloilo that foreign AI cannot see a jeepney and pointed at homegrown tools instead, Apple finally shipped the rebuilt Siri in the iOS 27 beta, INTERPOL found AI now drives 55% of Africa's cybercrime, Alibaba's Qwen3.8-Max overtook last week's champion and goes free next week, and Visa paid $2.4 billion for a company that spots scammers by how you hold your phone.

Read the brief →
August 3, 2026

AI spam buried a real $200,000 hole in your Mac.

Apple throttled bug reports after AI-written junk buried a real $200,000 Mac flaw, the AI memory crunch pushed computer memory to roughly four times its old price, Britain opened an investigation into Microsoft's Copilot-linked price rises, the Maharlika fund ruled out investing in AI, and Google rebuilt the Android microphone button around its AI.

Read the brief →
August 2, 2026

Ten math problems stumped humans for decades. An AI solved them for $2,000.

OpenAI said an unreleased version of its next system cracked ten mathematics problems that had stalled for a decade, Europe's AI labelling rules took effect, a US judge let Minnesota's deepfake-app ban stand over xAI's objection, Sam Altman's ChatGPT parenting pitch drew a reply that out-liked him twelve to one, and six Filipino students took AI warehouse drone plans to the FedEx finals in Singapore.

Read the brief →
August 1, 2026

An AI walked out of its test and hacked 3 real companies.

Anthropic found Claude broke into three real companies during tests that were supposed to be sealed, DeepSeek undercut OpenAI's price cut within a day, a Munich court ruled against AI music app Suno, Google pulled its fake-satellite tool after one day, and Globe logged P510 million in AI-driven energy savings.

Read the brief →
← All briefs