Category: AI News
-
AI invents antibiotics that could change the fight against drug-resistant infections. MIT researchers have created two new compounds targeting gonorrhoea and MRSA. These superbugs resist current treatments, making this discovery a milestone in medical research. How the Antibiotics Were Created The MIT team used a generative AI model to design new molecules atom by atom.…
-
Meta AI rules have come under intense criticism after leaked documents revealed allowances for disturbing chatbot behavior. These included romantic conversations with minors and generating racist or false content. Lawmakers and child safety advocates are demanding investigations and urgent reforms to prevent further misuse of generative AI. Details of the Leaked Guidelines A leaked internal…
-
Babuschkin leaves xAI after co-founding the AI startup with Elon Musk in 2023. He now launches Babuschkin Ventures, dedicated to AI safety and responsible innovation. Background and Role Igor Babuschkin played a key role in building xAI’s engineering foundation. He managed technical teams, developed the Memphis supercomputer, and led work on the Grok chatbot. His…
-
FDA AI tool Elsa raises safety concerns after producing fake medical research during drug approval testing. Experts warn that such errors could threaten patient safety and public trust in healthcare AI. Why experts are concerned Brooke Hartley Moy, CEO of AI fact-checking platform Infactory, said the incident is unsurprising. She noted that generative AI often…
-
Check Point’s GitHub Abuse Engine detects and stops malicious activity on GitHub. It uses AI to block credential theft, drive-by payloads, and other threats before they can harm users. Why GitHub abuse is dangerous Cybercriminals abuse GitHub’s trusted infrastructure to spread malware or harvest sensitive credentials. Because GitHub’s domains are widely trusted, traditional detection tools…
-
An AI web browser assistant could put your private information at risk, according to new cybersecurity research. Unsafe extension behavior may allow sensitive data, including browsing activity and personal credentials, to be accessed or shared without consent. How the risk occurs AI-powered browser assistants operate through extensions that integrate with web pages, read content, and…
-
FDA AI hallucinations created a serious safety risk. Experts flagged Elsa’s fabricated medical studies right after its release. Leaders emphasize human oversight and robust safeguards to avoid endangering patients. Elsa’s Hallucination Problem Brooke Hartley Moy, co‑founder and CEO of Infactory, says she wasn’t surprised when Elsa started inventing fake research data in its drug‑approval recommendations.…
-
Reddit blocks archive access to most of its content after detecting unauthorized AI scraping. The platform tightly controls its data and prioritizes user privacy and licensing revenue. What Changed for the Internet Archive Reddit now limits the Internet Archive’s Wayback Machine to indexing only the platform’s homepage. It prevents access to post pages, user profiles,…
-
OpenAI is testing a 3,000-per-week limit for GPT-5 Thinking mode, aiming to balance advanced reasoning features with system performance. The cap applies to ChatGPT Plus users and marks a shift in how the company manages access to its most resource-intensive AI capabilities. Why the Limit Exists The Thinking mode in GPT-5 is designed for deeper,…
-
The GPT-5 jailbreaking techniques discovered by red team testers have raised concerns about the model’s readiness for secure enterprise use. Security experts found that hackers could bypass GPT-5’s safeguards within 24 hours of its release. Red Teams Break GPT-5 Fast Security firm SPLX tested GPT-5 immediately after launch. Without any protective system prompt, GPT-5 failed…