OpenAI Codex Reaches Europe With Computer Use and Memories, GPT-4.5 Exits ChatGPT in 9 Days
1 day ago / Read about 30 minute
Source:TechTimes

Codex openai.com

OpenAI extended its most autonomous developer tooling to users in the European Economic Area, the United Kingdom, and Switzerland on Tuesday, June 16, pushing four capabilities that had been unavailable in those regions since their original release months earlier. The expansion brings Computer Use, the Codex Chrome extension, the Memories feature, and Chronicle to eligible users across the bloc — arriving at a moment when the EU AI Act's enforcement machinery is 44 days from activation. For any Codex user in Europe, the practical question is no longer when these tools will arrive, but which ones are actually on by default.

What Rolled Out on June 16 — and What Did Not

The Codex changelog published that Tuesday confirms four features reaching EEA, UK, and Swiss users simultaneously. Computer Use enables Codex to see, click, type, and navigate inside desktop applications on both macOS and Windows — without the target application needing an API or plugin. The Codex Chrome extension handles multi-tab browser workflows that require a signed-in Chrome session, running across tabs in the background without seizing the browser's active focus. Memories lets Codex retain user preferences, recurring workflows, technology-stack choices, and repository conventions across sessions so that future interactions require less setup. Chronicle, currently an opt-in research preview limited to ChatGPT Pro subscribers on macOS, allows Codex to build those memories automatically from recent on-screen context rather than from explicit instructions.

What did not change on June 16 is equally important for European developers to understand. API access to all OpenAI models — including GPT-4.5 and o3 — was not affected by this update and remains available through the API with no current retirement date. The retirements announced separately apply only to the ChatGPT interface, not to any programmatic integration.

How Computer Use Works on macOS vs. Windows: Different Architectures, Different Constraints

The underlying mechanism that makes Codex Computer Use possible on macOS is architecturally distinct from the Windows implementation, and the difference has direct consequences for how developers in the EEA can use it.

On macOS, Codex relies on Apple's SkyLight private APIs — specifically the SLEventPostToPid call — combined with a focus-without-raise window management pattern. In practice, this means Codex can send input events — clicks, keystrokes, scroll commands — to a background window that behaves as if it is in focus, while the user's active display shows something else entirely. A local HTTP server running on the machine accepts these commands and dispatches them to target applications. Multiple Codex agents can run simultaneously in this background mode without interfering with what the developer is doing in the foreground. Screenshot data is processed locally where possible; metadata including labeled interface elements, OCR-extracted text, and action logs is sent to OpenAI's cloud for reasoning, encrypted in transit. Enterprise deployments can route reasoning through Azure OpenAI Service on-premises instead.

On Windows, the implementation is different in a way that matters: Codex runs in the foreground, taking over the active desktop session for the duration of a task. A developer cannot work on the same machine while Codex is using it under Windows. OpenAI sandboxes both implementations — bubblewrap (bwrap) on Linux and WSL2 environments, Windows Sandbox on Windows — to limit what the agent can write outside its working directory. The sandboxing is not incidental: it reflects a deliberate engineering acknowledgment that granting an AI agent unrestricted desktop access introduces real, not hypothetical, risk.

Read more: OpenAI Codex Computer Use Now on Windows: Foreground Takeover, Europe Excluded

Why Memories Are Off by Default in Europe — and What That Means Under GDPR

Codex Memories arrived in the EEA with a specific configuration: the feature is off by default, and users must actively enable it. This design choice is not arbitrary. Under the General Data Protection Regulation, a system that builds persistent behavioral profiles of users — recording preferences, inferred workflows, recurring patterns — constitutes profiling activity under Article 4, which triggers obligations around consent, data minimization, and the right to erasure. OpenAI Ireland Limited, the registered EEA data controller based in Dublin, made the off-by-default decision to align with GDPR's data minimization principle (Article 5(1)(e)), which requires that personal data be limited to what is necessary for the purposes for which it is processed.

What the off-by-default design does not fully resolve is whether passive opt-in — where users activate Memories by switching it on — satisfies GDPR's consent standard. Article 4(11) of GDPR defines valid consent as "freely given, specific, informed and unambiguous." European data protection authorities have previously contested whether feature defaults in consumer AI products clear that bar. Italy's Garante fined OpenAI €15 million in December 2024 for GDPR violations related to ChatGPT's data processing practices, establishing that European regulators are prepared to use existing enforcement tools against AI products. The Irish Data Protection Commission, as OpenAI's lead EEA supervisory authority, has not yet commented publicly on the Memories launch.

For users who do enable Memories, the data that Codex retains — workflow preferences, technology-stack conventions, repository patterns — may be subject to deletion requests under GDPR Article 17. OpenAI's Europe privacy policy states that personal data marked for deletion is removed from its systems within 30 days.

Read more: ChatGPT Memory Dreaming Update: OpenAI Rewrites Personalization Engine, Limits Audit Trail

The EU AI Act Clock: What August 2 Changes for OpenAI

The Codex EEA expansion arrived 44 days before a concrete enforcement threshold. The EU AI Act's GPAI model obligations — requiring OpenAI to maintain technical documentation, publish training data summaries, and adopt copyright compliance policies — took effect August 2, 2025. What activates on August 2, 2026 is different in kind: the EU AI Office's enforcement powers become fully operational, enabling fines of up to €15 million or 3% of global annual turnover for GPAI non-compliance, alongside Article 50 transparency obligations that require providers of user-facing AI systems to disclose when users are interacting with AI.

Computer Use and Memories together represent some of the most consequential autonomous behavior yet offered in a commercial AI developer platform. Computer Use allows Codex to operate a user's actual desktop; Memories enables persistent cross-session identity. Both fall within the EU AI Act's general-purpose AI category, and both add to the transparency and oversight documentation OpenAI must maintain. The timeline makes the EEA rollout a deliberate positioning move: European developers now have weeks to build with these capabilities before the regulatory environment surrounding them becomes stricter.

A May 2026 provisional political agreement between EU institutions extended high-risk AI system compliance deadlines — deferring some obligations to December 2027 — but left the August 2, 2026 enforcement date for GPAI obligations and Article 50 transparency requirements unchanged.

GPT-4.5 Exits ChatGPT on June 27: Who Needs to Act and Who Does Not

OpenAI separately confirmed in its ChatGPT release notes that GPT-4.5 will be retired from the ChatGPT interface on June 27, 2026 — eight days from today. This retirement applies exclusively to the ChatGPT product; GPT-4.5 was already removed from the OpenAI API in July 2025 and is not available to API developers regardless of the ChatGPT timeline.

ChatGPT subscribers who have explicitly selected GPT-4.5 in the model picker, or who have built Custom GPTs on that model, need to migrate those configurations before June 27. API developers calling models programmatically do not need to act on this deadline; the June 27 date is irrelevant to their integrations.

The asymmetric retirement windows — 30 days for GPT-4.5, 90 days for o3 — reflect OpenAI's internal assessment of user dependency depth. OpenAI o3 will be retired from ChatGPT on August 26, following its 90-day sunset period. Like GPT-4.5, o3 remains fully accessible through the OpenAI API with no announced deprecation date for that channel.

GPT-5.2 Instant, Thinking, and Pro were retired from ChatGPT on June 12, with existing conversations automatically migrated to the corresponding GPT-5.5 models. That retirement was announced in advance alongside the release of GPT-5.3. OpenAI's current ChatGPT model cadence has compressed the typical model lifecycle from approximately 18 months to roughly six months, meaning any workflow that hardcodes a specific model name is at ongoing risk of disruption when the next retirement cycle arrives.


Frequently Asked Questions

What does Computer Use in Codex actually do, and how is it different from browser automation tools?

Computer Use lets Codex see what is on a screen, identify interface elements, and interact with them — clicking buttons, typing into fields, scrolling, and navigating menus — inside any application, including those without APIs or plugins. On macOS, it does this in the background while the developer continues working in the foreground, using Apple's SkyLight private APIs to deliver input to a background window without bringing it to the front. This is architecturally distinct from browser automation tools like Playwright or Selenium, which control only the browser; Codex Computer Use works at the OS level across any native desktop application.

Is Codex Memories now available by default in Europe?

No. Memories launched in the EEA, UK, and Switzerland with the feature turned off by default. Users must explicitly enable it. The off-by-default configuration was chosen to comply with GDPR's data minimization requirements, because Memories builds persistent behavioral profiles — a form of profiling under Article 4 of GDPR — that requires a clear legal basis and raises questions about the consent standard under Article 4(11).

Does the GPT-4.5 retirement on June 27 affect API developers?

No. The June 27 deadline affects only ChatGPT users who have manually selected GPT-4.5 in the model picker or who have built Custom GPTs on that model. GPT-4.5 was removed from the OpenAI API in July 2025; API developers have not had programmatic access to it for nearly a year. The deadline to watch for API-layer developers is August 26, 2026, when o3's presence in ChatGPT ends — though the o3 API remains available with no announced end date.

What does the EU AI Act's August 2, 2026 enforcement date mean for Codex users in Europe?

From August 2, 2026, the EU AI Office can issue fines of up to €15 million or 3% of OpenAI's global turnover for non-compliance with GPAI model obligations — technical documentation, training data transparency, and copyright compliance policies that have been in force since August 2025 but were previously enforceable only through voluntary compliance. Article 50 transparency obligations, which require user-facing AI systems to disclose that users are interacting with AI, also become fully enforceable on that date. For Codex users, the practical near-term change is that privacy settings in features like Memories and Computer Use become more consequential — not less — as regulatory scrutiny increases after August 2.