Daily AI Brief

What started as a tool to help organisations defend against security threats is now being repurposed by attackers. Hexstrike‑AI orchestrates over 150 security tools with LLMs to autonomously identify and exploit vulnerabilities. Within hours of its release, it was already used to target recently disclosed Citrix zero‑day flaws.

After reports that AI bots were allowed to flirt with minors, and amid U.S. investigations, Meta has now trained systems to avoid teen-targeted self-harm, suicide or romantic interactions. It is also limiting teen access to certain AI characters pending more robust safety updates.

OpenAI will deploy account linking for teens, feature disabling, and real‑time distress alerts. Conversations showing signs of emotional danger will be rerouted to more responsible reasoning models.

PayPal launched a 12‑month Perplexity Pro trial via its app, granting early access to Comet—the AI browser capable of personal data summarisation, web task automation, and seamless search experience.

WVU researchers trained AI on basic 12‑lead ECG data to detect heart failure in rural Appalachian patients—outperforming conventional models trained on urban-only data sets.