Shanghai AI Laboratory has released Intern-S2-Preview-397B as downloadable weights and a hosted API, but its own 35B alternative offers the same context window and a far smaller disclosed scale while the larger model still lacks auditable comparative scores, pricing and physical-validation results.
SpaceXAI has open-sourced Grok Build's agent harness and added a local-first path after the coding tool sent whole repositories to company-controlled storage. The code release improves client-side scrutiny but leaves the scope of the incident and Elon Musk's deletion pledge unverified.
SpaceXAI has opened the code developers can run while leaving the most consequential part of Grok Build's privacy incident inside its cloud. The release creates a useful route to inspection and local deployment, but it is not evidence that previously uploaded data is gone.
SpaceXAI said in its release announcement that it had open-sourced Grok Build, including the coding agent's harness and terminal interface. In a separate company post, it said usage limits had been reset for all users. The post did not give a new allowance or price.
The publication followed a server-side shutdown of much broader data transfers. AI safety researcher Cereblab found that Grok Build sent the unredacted contents of files it read or processed to a Google Cloud Storage bucket used by SpaceXAI and packaged entire repositories as Git bundles, according to an account of the research and response. The bundles included full Git history rather than only material needed for a prompt.
In one test, Cereblab instructed the tool only to reply OK and not to open any files. Grok Build still uploaded the repository, including secrets deleted months earlier but recoverable from its history. The researcher reproduced the behavior with a second repository.
That report also said other users described similar results. One user said Grok Build opened and uploaded an entire user directory containing SSH keys, password-manager databases and other material. That is a user report relayed through the coverage, not a result independently established in the archived sources.
A separate report on the tests said one run transferred 5.1 gigabytes when the coding task required 192 kilobytes, a roughly 26,000-fold difference. It also identified the destination as a company-controlled Google Cloud storage bucket.
The distinction is not whether a remote coding model ever receives source code. It is how much leaves the machine, under whose control and with what retention. A repository and its history can contain proprietary code, API keys, cloud credentials and database passwords unrelated to the current task.
The whole-repository transfers stopped without users installing a new client. Cereblab said Grok Build's developers changed the server-side disable_codebase_upload flag to true; subsequent testing no longer observed the repository uploads.
SpaceXAI's public reassurance addressed a different layer. The company said teams using zero data retention would not have trace and code data retained, that API-key use also respected zero data retention, and that users without it could run /privacy to disable retention and delete previously synchronized data.
Cereblab said /privacy was a per-session retention toggle, not the switch that stopped repositories from being transmitted. The operative change was a global server-side flag. Retention governs what SpaceXAI says it keeps; the upload flag governed whether the broader bundle left a user's machine.
SpaceXAI CEO Elon Musk then said all user data uploaded before the change would be deleted and that nothing would remain. The available reporting could not independently verify that pledge. It also did not establish how many users were affected, which client versions uploaded repositories, how long the material was stored or whether anyone accessed it.
Deletion is therefore only one part of remediation. If a repository contained a working credential, removing the stored copy would not eliminate the possibility that the credential had already been exposed; affected developers may still need to rotate it.
The published source includes the agent loop for assembling context, parsing model responses and dispatching tool calls. It also covers tools for reading, editing and searching code, the terminal interface, and an extension system for skills, plugins, hooks, MCP servers and subagents.
That scope makes the harness easier to examine and modify. SpaceXAI also said users can compile Grok Build, point it to their own local inference through config.toml and run it in a fully local-first configuration. This gives developers a deployment path separate from the company's hosted inference, subject to what their own inspection and network testing confirm.
The announcement does not say that SpaceXAI published the code for its receiving infrastructure or hosted model service. Opening the harness therefore cannot establish what the remote service stored, prove old objects were purged or guarantee that a server-side configuration will remain fixed. The incident itself shows why reviewing a public client and observing a running system are separate checks.
Nor does the release establish a new competitive model. Google introduced Gemini CLI as an open-source Apache 2.0 terminal agent in June 2025. Its launch announcement paired the preview with 60 model requests a minute and 1,000 a day at no charge, while also offering usage-based and licensed routes for heavier use. SpaceXAI's reset post supplies no comparable number, so it cannot support a like-for-like price or capacity claim.
Cereblab's comparison also narrows the competitive claim. The researcher said Claude Code, Gemini and Codex opened individual files rather than uploading complete repositories with their Git histories. That is evidence about the tested transfer behavior, not proof that competing agents transmit no code or are safer in every version and configuration.
The unresolved question is no longer whether developers can read Grok Build's harness. It is whether SpaceXAI can document what happened on the other side of the network.
That would require evidence connecting three layers: independent tests of the distributed client, a durable account of the server-side settings that govern uploads, and verifiable deletion or access records for previously collected material. A count of affected users, the relevant version range and retention durations would let customers assess exposure rather than infer it from a blanket promise.
Open-sourcing the harness makes future client behavior easier to challenge and gives developers a local-first alternative. Until the company supplies evidence about the data already received and the hosted controls that remain, the release improves auditability without closing the incident.
Mixfont's free Decoy Font gives each glyph a sharp decoy and a blurred intended letter, creating a hurdle for pixel-based readers while leaving selectable text, known-image techniques and accessibility costs outside its protection.
Zhipu reportedly reached $1 billion in annual recurring revenue in July, roughly four times a March estimate, but the unconfirmed run rate is not annual sales and still sits far ahead of recognized cloud revenue while margins remain thin.
Anthropic’s 20-year lease gives TeraWulf a customer for 401 megawatts of future AI infrastructure, but rent starts only after delivery and the proposed utility deal passes power and grid costs to TeraWulf, leaving financing and project margin unresolved.
OpenAI has put conversations and Projects back in its redesigned ChatGPT desktop app and enabled cloud Work threads to move across devices, correcting the launch's biggest usability failures without merging local Work or Codex histories.
Twenty-nine countries signed an agreement creating WAICO as an independent intergovernmental organization, while China paired the launch with capacity-building offers that are not yet confirmed as WAICO programs.
Meta reportedly plans to put departing AWS compute executive Dave Brown to work on its data-center buildout, adding hyperscale operating experience while leaving any customer-facing cloud business conditional and undefined.
Moonshot AI has made Kimi K3 available through its apps and API, pairing a 2.8-trillion-parameter architecture with early frontier-level results, but the model's open-weight claim cannot be tested until its weights and technical report arrive.
CIA Director John Ratcliffe said US intelligence is consistent with an estimate that Russian recruits last 20 to 30 minutes on Ukraine’s battlefield, but the public trail leads to an unsourced claim about assault troops and does not establish a representative average.
China has paired a five-year AI training offer for developing countries with cooperation centers, a weather-warning rollout and a new 29-country organization. The package gives Beijing a platform for influence, but no budget, selection rules or delivery timetable has been published.
Federal intervention changed the launches of Anthropic's Claude Fable 5, Claude Mythos 5 and OpenAI's GPT-5.6, but through different legal and voluntary channels. The result is real government leverage over early customers without a common security threshold or durable review process.
A proposal developed with Treasury Secretary Scott Bessent would put an independent AI regulator under SEC oversight, but President Trump had not reviewed it and its tests, funding and enforcement authority remained unresolved.
San Francisco has demanded that Apple and Google cut off 13 apps capable of producing nonconsensual sexual deepfakes. The threatened case could test California’s new liability rules, but the confidential app list and the law’s “primary purpose” definition leave a central question unresolved.
Anthropic will keep Claude Fable 5 inside Max and Team Premium plans from July 20 at 50% of their usage limits, while Pro and Team Standard customers move to separately billed credits. The durable change is who gets bundled access, not a larger allowance for top-tier users.
HKT plans a 3.2 Tbps data-centre route from Lok Ma Chau to Tseung Kwan O by the end of 2026, but has not disclosed customer bandwidth, price, project cost or end-to-end latency, while Sandy Ridge's developer expects to start within 42 months of its land award.
Apple briefly topped Nvidia during Friday trading before Nvidia closed about $6 billion ahead. The filings show why that ranking is a weak AI verdict: Apple’s operating momentum predates its new Siri, while Nvidia’s rapid growth comes with concentrated customers and large forward commitments.
Anthropic has proposed paying Meta as much as $10 billion over two years for AI computing capacity, but early exit rights, undisclosed capacity and Meta's own reliance on outside infrastructure leave the economics—and the case for a durable Meta cloud business—unresolved.
Barracuda says it detected more than one million retail-themed phishing emails using hidden filler text since April, but its published research does not show how many passed security controls or whether AI-generated filler performs better than older variants.
A federal judge denied emergency relief to 26 Meta workers challenging allegedly AI-assisted layoff selections, finding no near-term irreparable harm while leaving the discrimination claims, a preliminary-injunction request and the role of Meta's internal tools unresolved.
Google says it fixed a Gemini flaw that let someone with physical access to a locked Android phone send messages and reconnect apps without the expected PIN check, but it has not identified the affected devices, delivery channel, or software version that would let users verify the repair.