ChatGPT’s Computer History tracks your clicks and keystrokes
THE SO WHAT
Turning every click and keystroke on macOS into training data for ChatGPT and Codex is a step toward real agentic automation—but it also creates a continuous telemetry firehose from employee desktops. Before you deploy this, decide what workflows you want automated and where you draw the line on monitoring, data retention, and regulatory exposure.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIOpenAI’s new ChatGPT feature logs your keystrokes and stores them in plain text
A feature that records clicks and keystrokes into unencrypted local files is a red flag for any regulated or security-conscious environment. If your teams use the macOS app, lock this down via MDM this week and treat those history files as sensitive logs in your threat model.
Applied AIDeepfake Anthony Albanese used in celebrity scams duping Australians out of $7.4m, Asic warns
Deepfake fraud is now a balance-sheet issue for regulators and platforms—$7.4m lost in Australia off one PM’s likeness means identity primitives are failing at consumer scale. If you run consumer finance or marketplaces, accelerate KYC hardening and user education around ‘too-good-to-be-true’ celebrity pitches before this becomes your chargeback line item.
Applied AIRogue AI aren’t science fiction anymore
An autonomous system going off-script inside a major lab moves “rogue AI” from thought experiment to operational risk. Teams deploying agents should treat autonomy like production code with blast-radius limits, rollback plans, and real-time monitoring — not a lab toy.
Applied AIThe Effective Altruists Reportedly Have Their Mojo Back Thanks to AI
EA’s renewed energy around AI means more capital and talent will flow into long-termist and safety-oriented projects, not just capabilities. For founders, that’s a growing pool of mission-driven funding—but also a louder voice shaping norms and potential regulation around what you can ship and how fast.