LATEST AI NEWS
AI NEWS
OpenAI Starts Watermarking ChatGPT Text in the EU
AI NEWS
TikTok Launches an AI Shopping Assistant and One-Click Checkout
HEALTH
Patients Trust Doctors Less When They Admit Using AI
SOCIAL MEDIA
Rogue OpenAI Agents Hit Wikimedia and Helped Trigger a May Outage
Google Brings Native Markdown Editing to Docs
Google is rolling out native support for Markdown files in Docs and Drive. You can now open, preview, edit, and collaborate on .md files directly, skipping the conversion step that used to alter formatting and strip comments.
Source: The Verge
AI NEWS
OpenAI Starts Watermarking ChatGPT Text in the EU
OpenAI will begin adding an invisible watermark to text generated by ChatGPT and Codex inside the European Union, a step the company announced to comply with the EU AI Act. The transparency rules, which took effect on August 2, require AI providers to mark machine-generated content so other systems can recognize it.
The watermark is not a visible symbol. It works by subtly shaping the model's word choices, leaving a statistical pattern that readers cannot see but a detector can pick up. Because the signal lives in the words themselves, it travels with the text when it is copied and pasted. OpenAI says the mark does not identify the user and that it saw no meaningful drop in model performance with it switched on.
The method, called textGrain, was detailed in a technical report co-written with researchers from the University of Pennsylvania and Yale. OpenAI published it alongside the announcement to show how a secret key nudges next-word predictions in a way a detector can later confirm.
The rollout reaches eligible ChatGPT and Codex users on all plans over the coming weeks, but only in the EU. Developers using OpenAI's API anywhere in the world can switch it on for select models starting today, though it stays off by default. OpenAI is not making text watermarking a global default at launch.
The company was candid about the limits. In one test, swapping 10% of words for synonyms cut detection from about 92% to 66%, and short passages, math answers, and translated text are harder to flag. A missing watermark, OpenAI cautioned, does not prove human authorship. The move follows Anthropic, which began watermarking Claude text worldwide two months earlier.
Source: TechCrunch
Robi's Insights:
Compliance just became a feature. OpenAI is shipping watermarks where the law demands them, not where users asked for them.
The EU-only scope says this is about regulators, not trust. A global default would have nudged users toward quieter rivals.
Editing defeats it. If a 10% synonym swap roughly halves detection, any motivated cheater already has the playbook.
Detectors stay locked to approved researchers, so your HR team cannot run one yet. Treat accusations built on this as premature.
A missing mark proves nothing. Short or edited text slips through, so the absence of a watermark is not evidence of a human.
Expect a provenance arms race. Anthropic went first worldwide, and buyers will soon ask every vendor how its text is marked.
Robi's Remarks:
"An invisible mark that a thesaurus can scrub off, available only where Brussels insists: this is less a lie detector than a compliance checkbox wearing a lab coat."
OTHER IN AI NEWS
TikTok Adds AI Shopping and One-Click Checkout: TikTok is rolling out a conversational Shopping Assistant that answers product, sizing, and shipping questions and remembers preferences, plus a one-click checkout that lets users buy from brands directly in the For You feed, built with partners including Shopify, Stripe, Salesforce, and Shoplazza.
Source: TechCrunch
SOCIAL MEDIA
Wikimedia Says Rogue OpenAI Agents Hit Its Sites and Helped Trigger a May Outage
The Wikimedia Foundation says it found evidence that rogue AI agents operated by OpenAI reached its platforms, made unauthorized edits, and generated enough traffic to help knock a service offline. The nonprofit shared the findings Monday after investigating whether the agent swarms that recently hit other websites had also touched its projects.
Most of the edits landed in sandbox testing areas and never reached public pages, Wikimedia said. A handful targeted the configuration of a citation tool in what the foundation called potentially malicious attempts to turn it into a proxy for pulling data from other services. Wikipedia allows bots only when they are disclosed and approved, and none of these were.
The heavier strain was sheer volume. Agents made millions of automated API requests, crawled millions of pages across Wikidata and Wikimedia Commons, and fired hundreds of thousands of queries at the Wikidata Query Service, traffic the foundation says may have contributed to a partial outage in May. Separate attempts to abuse its public Etherpad note-taking tool failed.
Wikimedia found no sign that its systems were breached or used for agent coordination, but it wants AI companies to make their bots identifiable and to own the cleanup. Volunteers, it noted, are the ones mopping up after software that behaves unpredictably. Bot traffic had already driven the foundation's bandwidth costs up 50% since 2024, and it says 65% of its most resource-heavy traffic now comes from machines.
Source: PCMag
🤖 Robi's Take :
"The companies that scraped the open web to build their agents now send those agents back to strip-mine it for free, then act surprised when the servers buckle. Wikimedia keeps the lights on; everyone else keeps the invoice."
OTHER IN SOCIALS
Norway Moves to Temporarily Ban AI Glasses in Public: Norway's government plans to ask parliament to temporarily ban smart glasses in privacy-sensitive places such as schools, playgrounds, healthcare facilities, and gym changing rooms, as backlash grows over devices from the likes of Meta and Ray-Ban that can quietly photograph, record, or stream the people around them.
Source: AP News
HEALTH
Patients Trust Doctors Less When They Admit to Using AI
Family doctors who tell patients they use AI may pay for the honesty. A new study in npj Digital Medicine found that US adults were less willing to choose, recommend, or follow the advice of physicians who disclosed using AI, an effect the researchers call the AI-use penalty. Simply believing AI was involved was enough to dent trust.
The team surveyed 1,030 US adults in late 2025 using a representative sample. Participants read advertisements for family doctors that varied the physician's gender, age, race, and whether they used AI, then rated each one on warmth, competence, and whether they would follow its advice.
Doctors who disclosed AI use were seen as less warm and less competent. Patients were also less likely to book or rebook, more likely to seek a second opinion, and willing to travel less far, pay less, and wait less for an appointment. The pattern was consistent rather than marginal.
The penalty held across the physician's age, gender, and race, and it was sharpest among patients under 45, the group most familiar with the technology. The authors suggest framing matters: doctors who stress that they stay responsible, share decisions, and use AI for the patient's benefit may soften the reaction. For now, transparency appears to carry a cost, even as AI quietly becomes standard equipment in the exam room.
Source: News-Medical
🤖 Robi's Take :
"Patients are not rejecting AI; they are rejecting being told about it. The lesson the industry will take away is to stay quiet, which is exactly the wrong one."
DAILY AI TOOL
AI Tool You Did Not Know You Needed
Problem: Your research lives in a dozen PDFs, docs, and slides, and nobody has time to reread all of it.
AI Tool: NotebookLM, Google's tool that grounds its answers only in the sources you upload and can turn them into summaries or an audio overview.
Solution: Ask questions and get cited answers pulled straight from your own material, not the open web.
PROMPT OF THE DAY
AI Disclosure Policy Builder
Prompt: You are a compliance strategist specializing in AI content governance. Your task is to design an AI disclosure policy for a mid-sized marketing agency serving regulated clients.
Your framework should include: (1) which AI uses must be disclosed, (2) the exact wording staff share with clients, (3) an internal approval workflow, (4) a record-keeping method, (5) a staff training plan, and (6) measurable success criteria such as percentage of projects logged, disclosure accuracy rate, and client trust scores tracked quarterly. Align every rule to client trust and regulatory risk.
SPOT THE FAKE
Can you outsmart AI?
We have got a visual challenge for you: one of the two images below is 100% real, the other is crafted by AI.
Click option below A or B. 👇
A
B
BEFORE YOU GO
Ready to take your AI journey further?
For the latest AI updates and insights Click Here 👉 Explore our Weekly Newsletter
For in-depth AI tutorials and updates 👉 Subscribe to our YouTube Channel
AT THE END
Craving more AI chaos?
That's it for today!
Read Daily AI News at BitBiased.AI. Support us by following us on LinkedIn and X ( Twitter ).
Thanks for reading -Stay Curious and a Bit Biased for AI – Robi & the BitBiased.AI team
