LIVE · cybersecurity feed
Live wire
ai

OpenAI ditches Recall-style screenshot surveillance for friendly keylogging

OpenAI has introduced a new feature called Computer History for its ChatGPT desktop application on macOS, available to ChatGPT Pro, Business, and Enterprise subscribers. This opt-in system records user interactions across various applications and websites, compiling them into "memories" organized on a timeline. The company states that this feature is designed to enhance ChatGPT responses,…

ZeroDay News ·

Source: The Register — Security

OpenAI has introduced a new feature called Computer History for its ChatGPT desktop application on macOS, available to ChatGPT Pro, Business, and Enterprise subscribers. This opt-in system records user interactions across various applications and websites, compiling them into "memories" organized on a timeline. The company states that this feature is designed to enhance ChatGPT responses, identify automation opportunities, and facilitate the resumption of previous work.

Computer History operates by capturing input events such as clicks, typing, keyboard shortcuts, and application switches, along with contextual information exposed by macOS accessibility features. These events are periodically summarized into text and stored in local memory files. OpenAI explicitly notes that the system does not capture screen images, microphone input, or system audio, nor does it record activity in private browsing modes.

This new feature replaces an earlier iteration, Chronicle, which reportedly used screenshots to compile similar histories. While Computer History avoids screenshots, it functions as a keylogging and event capture system. OpenAI's documentation includes a warning that "Computer History files can contain sensitive information. They are not encrypted by Computer History, and other programs running as your macOS user may be able to access them."

Interaction event files are intended to be saved locally for up to 48 hours before being deleted by ChatGPT and Codex. However, events are transmitted to OpenAI servers for the purpose of generating memories, which may be stored locally for longer durations and could be utilized in future chats that are subsequently sent back to OpenAI. The company asserts that it does not retain these event files after processing, unless legally mandated, and does not use them for training purposes. Despite this, OpenAI has previously provided chat logs in response to legal requests.

Computer History is currently unavailable in the European Economic Area (EEA), Switzerland, or the United Kingdom, and cannot be used by those accessing ChatGPT via API key or Amazon Bedrock. For Pro users, the feature can be enabled individually, while Business and Enterprise users require administrator approval.

OpenAI advises users to exercise caution, recommending that Computer History be turned off during communications with others unless prior express consent has been obtained. The company also suggests pausing the service or excluding applications that handle sensitive health, financial, or personal information.

The use of Computer History incurs additional costs due to the consumption of tokens during the summarization of activities and the creation of memory data. Furthermore, OpenAI acknowledges that the feature expands the potential attack surface for prompt injection. The company warns that "Computer History increases the risk of prompt injection from content in apps and websites," citing the example of ChatGPT or Codex potentially following malicious instructions encountered on a website.

ai
ShareXLinkedInWhatsAppFacebook

More News

view all →
breach

Criminal IP Introduces AITEM as the Next Evolution of Attack Surface Management

AI SPERA, the company behind the cyber threat intelligence platform Criminal IP, has announced AITEM (AI-Powered Threat Exposure Management), an expansion of its Attack Surface Management (ASM) offerings. AITEM aims to move beyond traditional asset discovery to provide a more comprehensive approach to understanding, prioritizing, and responding to security exposures. The platform is scheduled…

ai

AI Is Getting Really Good at Messing With Cybercriminals

Anti-cybercrime initiatives are increasingly deploying artificial intelligence to disrupt scammers by engaging them with lifelike bots, which the criminals mistake for genuine victims. This strategy aims to waste scammers' time and resources while gathering intelligence on their operations.

ai

The Third-Party Agent Problem: Why Security Built for AI You Chose Misses the Agents You Didn't

A new report highlights a significant blind spot in enterprise security architectures concerning the proliferation of third-party AI agents. The forthcoming 2026 State of Agent Security Report indicates that current security models, often designed to protect AI systems directly chosen and deployed by an organization, frequently fail to account for AI capabilities embedded within third-party…

security

Canadian cybersecurity executive arrested in federal extortion case

A Canadian cybersecurity executive was arrested in Pennsylvania on Thursday, facing federal charges of conspiracy to commit extortion. Edward Dubrovsky, 54, a co-founder and former Chief Operating Officer of CYPFER, a firm specializing in ransomware negotiation, is charged with conspiring to threaten the confidentiality of information for extortion and conspiring to commit Hobbs Act extortion,…

ransomware

FBI Arrests Ransomware Negotiation Firm Co-Founder in ShinyHunters Probe

The Federal Bureau of Investigation (FBI) has reportedly arrested Edward Dubrovsky, a co-founder of the ransomware negotiation firm Cypfer and currently associated with CyberSteward. The arrest is said to be part of the ongoing ShinyHunters investigation, specifically in connection with an incident involving FBI job applicant data.

security

Insider Cyber Extortion Plot Against Industrial Firm Lands Engineer in Prison

A former core infrastructure engineer has been sentenced to prison for an attempted cyber extortion plot against an industrial firm. The engineer reportedly deleted administrative accounts and reset hundreds of user passwords, subsequently demanding a ransom of 20 Bitcoin to prevent further disruption to the company's servers.