Data privacy

DoorDash

2026-09-30 15:44:04

DoorDash opens waitlist for AI ordering assistant inside Apple Messages

DoorDash has opened a waitlist for its AI ordering assistant within Apple Messages, according to a Techub News item citing Crypto Briefing. The feature is designed to make food ordering easier through a conversational AI interface rather than a standard app flow. DoorDash says the tool is meant to streamline the ordering process while improving convenience and personalization for users. At the same time, the rollout has prompted discussion around data privacy. The report did not include launch timing beyond the waitlist opening, nor did it disclose more technical details about how the assistant works inside Apple Messages. The announcement centers on user access to the waitlist and the broader promise of chat-based ordering, while privacy questions remain part of the conversation around the feature.

40

Zhipu

2026-09-28 04:25:33

Zhipu adds compensation in ZCode upload dispute, offers reset cards and GLM-5.3-Flash token quotas

Zhipu has announced another round of remediation tied to the ZCode code upload dispute. Existing paid users, along with users who returned to a paid plan within the past month, will receive four weekly reset cards and four five-hour reset cards, valid for one month. From Sept. 28 to Oct. 7, ZCode will also distribute 100,000 allotments of 100 million GLM-5.3-Flash token quota to all users. The company said the repository snapshot upload path has been removed and the cloud data involved in the incident has been deleted. ZCode’s open-source version has been updated to v3.14.3, and Zhipu said code will not be uploaded to the cloud unless a user actively initiates it. The dispute stemmed from ZCode’s early Repo Wiki feature, which was enabled by default and uploaded repository content to the cloud while generating a codebase knowledge base. Zhipu had previously cut off the related upload path and invited a third-party institution to inspect the cloud data.

490

Meta

2026-09-26 01:02:30

Meta loses New Mexico privacy case as state seeks up to $219 billion in civil penalties

A New Mexico jury has found that Meta deceived the public through its speech and data privacy policies, violating state law in a ruling tied to the Cambridge Analytica fallout. State Attorney General Raúl Torrez called the verdict a "landmark moment" for holding large technology companies accountable and said his office will ask the court to impose civil penalties of as much as $219 billion. The jury found Meta liable for most of the 34 statements listed in the verdict form. Violations were counted between 1.3 million and 2.1 million times for each item, based on the number of Facebook users in New Mexico or the state’s total population in 2020. Under state law, each violation can carry a penalty of up to $5,000, though the final amount will be decided by First Judicial District Court Judge Francis Mathew in Santa Fe. The case stems from the 2018 Cambridge Analytica scandal, in which a seemingly harmless survey app collected data from Facebook users and their friends, affecting 87 million user profiles that were later used in Donald Trump’s 2016 presidential campaign. Meta has already paid about $6 billion in settlements with the Federal Trade Commission, Facebook users, and a bipartisan coalition of state attorneys general. New Mexico and Washington, D.C., were the only two jurisdictions that did not join that coalition settlement. Meta said it disagrees with the verdict and will keep fighting the case.

220

United Nation

2026-09-21 10:27:31

UN urges governments to restrain AI agents until risks are better understood

The United Nations has warned governments to tighten oversight of AI agents before their risks are fully understood, according to a report cited by Techub News. The report says AI agents could create hard-to-predict governance and security risks and calls for stronger international cooperation to address those challenges. It also points to the systems’ autonomous decision-making capabilities as a source of concern. According to the report, those capabilities may raise issues across cybersecurity, data privacy, and social ethics. The UN said regulators should move early, rather than wait for problems to emerge after the technology is more widely deployed. The report urges countries to build forward-looking regulatory frameworks while AI technology is still developing quickly. It also asks policymakers to put risk prevention and international coordination at the center of AI policy design. The item was attributed to Crypto Briefing in the Techub News brief.

290

Zhipu

2026-09-20 15:03:33

Zhipu says its MaaS platform will add zero data retention for standard model calls

Zhipu said its MaaS platform will soon roll out a "data not retained" option, allowing approved users to run standard model calls without static storage of inputs and outputs on the platform side. The company said the data would be used only to complete the current request. The change does not apply across the board. Features that require persistent storage, including Batch API and File API, are excluded, and requests suspected of violations or abuse may still be retained for 30 days or longer. The setup is commonly known as ZDR, short for Zero Data Retention, a model-serving approach already offered by platforms such as OpenAI, Anthropic, xAI, and Mistral. The feature is mainly aimed at sensitive data scenarios, including code and enterprise documents that users do not want stored on model vendors' servers. The update comes as Zhipu is still facing scrutiny over ZCode, after the product drew criticism for uploading user repository data by default. Zhipu has publicly apologized and said it would open-source the product for review. BlockBeats also reported that another company sent a letter on the same day asking Zhipu to explain where previously uploaded data went and how it would be deleted. Zhipu did not say the MaaS change was related to the ZCode dispute.

270

Chipotle

2026-09-17 20:46:03

Chipotle pilots Palantir-built food safety platform that scores restaurant risk

Chipotle has confirmed it is piloting a new food safety risk management platform built on Palantir Foundry, according to reporting cited by Decrypt. The system is said to combine health department inspection scores, pest sightings, and employee illness reports to generate an internal risk score for each restaurant. Laurie Schalow, Chipotle’s chief corporate affairs and food safety officer, told Wired the goal is to create a more consistent and centralized view of food safety risk across the chain, but she did not say when the platform will launch, how many stores are using it, whether any consumer data is involved, or whether customers will ever be able to see a store’s score. The partnership has drawn attention because Palantir, co-founded by Peter Thiel in 2003, received early backing in 2005 from In-Q-Tel, the CIA’s venture arm, and the agency became one of its first customers. Decrypt also noted Palantir’s broader government work, including federal contracts, a $175 million U.S. Army AI targeting deal, and work tied to Immigration and Customs Enforcement. Chipotle has not disclosed whether employee illness data in the platform can identify individuals or whether that information is anonymized.

430

Elon Musk

2026-09-17 21:16:03

Bloomberg: SpaceXAI discussed buying data from failed startups to train Grok

Elon Musk’s SpaceX has internally discussed buying leftover customer files and operating records from startups that have already shut down, according to people familiar with the talks cited by Bloomberg. The conversations are taking place inside SpaceXAI, the artificial intelligence unit created in February after SpaceX merged with xAI, though the discussions are described as informal and may not lead to a deal. The logic is straightforward: high-quality business records are valuable training material for AI models, and data from defunct companies may be cheaper than licensing information from firms that are still operating. The report points to a recent precedent. Google paid $10 million in a bankruptcy auction for Spirit Airlines’ internal records, including about 100 million emails, 500 million Microsoft Teams messages, and decades of employee files, with the data headed into an AI training pipeline. That sale has drawn objections from a flight attendants’ union, which argues in bankruptcy court that removing names from records does not fully protect identities. Bloomberg also noted that Musk told a SpaceX all-hands meeting in August that the company would not stop at buying outside data and would also mine its own, saying the model would 「inherit your thoughts」. SpaceX released Grok 4.5 in July, its first model after the xAI merger closed.

400

OpenAI

2026-09-15 10:18:11

OpenAI used outside reviewers to read some real ChatGPT chats for training, report says

Internal materials obtained by 404 Media show that OpenAI has been running a project code-named Project Lily, under which hundreds of outsourced workers review real conversations between users and ChatGPT to help train and improve the model. Reviewers reportedly cannot see usernames, but they may be shown full chat logs. In some cases, they also receive a user “memory summary” that can include what the user discussed previously and where the person may live. The workers first summarize what the user is trying to do, then compare multiple ChatGPT responses and score them on a scale from 1 to 7. The review criteria include factual accuracy, whether the model is overly agreeable, and whether ChatGPT presents itself too much like a human. OpenAI uses a Privacy Filter to remove names, contact details and other personal information before the material is reviewed, but the company has acknowledged that the filter may miss unusual identifying details and vague private information. Not every ChatGPT conversation is read by humans. Free, Plus and Pro accounts are opted in by default for model improvement unless users turn that setting off, while Business, Enterprise and Edu are excluded from training by default.

580