OpenAI ditches Recall-style screenshot surveillance for friendly keylogging

OpenAI’s new opt-in macOS feature, “Computer History” for ChatGPT Pro/Business/Enterprise, records user interaction events (e.g., clicks, typing, app switches) as locally stored timeline memories, but the files are not encrypted and can contain sensitive data. The company warns it increases prompt-injection risk and that events are briefly local (up to 48 hours) before being processed/sent to OpenAI servers for memory generation. Availability is limited (not in the EEA/UK/Switzerland and not via API key/Bedrock), signaling a privacy-and-security tradeoff rather than a direct product performance upgrade.

Analysis

Near term, this is not a revenue event for AMZN; the market impact is almost entirely narrative. The real mechanism is procurement behavior: the more AI products resemble surveillance tools, the more enterprise buyers will demand auditability, data isolation, and admin controls — areas where AWS can claim structural advantage over consumer-first AI workflows. That makes this a modest relative positive for Bedrock and adjacent AWS services, but only if customers shift from front-end copilots to governed infrastructure.

The second-order risk is slower enterprise adoption of agentic desktop tools across the sector. Privacy/legal teams will likely lengthen approval cycles over the next 1-3 months, which could compress usage growth for consumer AI apps and push customers toward cloud-hosted, policy-controlled deployments. For AMZN, that is supportive only if it captures the migration; if adoption stays with proprietary assistants, AWS sees little direct benefit.

Contrarian view: the consensus may be overpricing the scandal risk and underpricing user inertia. Most enterprises do not react to privacy optics first; they react to whether IT can control retention, access, and logging. If OpenAI can prove low-friction governance, the headline fades quickly. The thesis is falsified if AWS commentary over the next 1-2 quarters shows no pickup in AI workload mix or if enterprise buyers keep choosing embedded assistant stacks despite the privacy concerns.

Key Decisions for Investors

• No immediate AMZN trade: the signal is too indirect for a high-conviction catalyst position.
• Add AMZN to a 3-6 month watchlist for a dip-buy if it sells off 3-5% on broader AI/privacy de-risking; the thesis is AWS as the governance beneficiary.
• If privacy scrutiny escalates into enterprise procurement delays, consider a relative long AMZN / short IGV pair over 1-3 months to express 'safe AI infrastructure' versus software governance risk.
• Do not chase options premium here unless AWS management explicitly quantifies Bedrock uptake; implied vol is likely to overstate the news flow.