The latest on Grok and xAI: model releases, product features, API and enterprise moves, government and policy coverage, and Elon Musk's broader AI strategy.
Topic brief
Brief updated Aug 6, 2026
Grok is the AI model family built by xAI, the company Elon Musk founded that has since been absorbed into SpaceX and is now marketed under the SpaceXAI name. Grok ships across X, a standalone app, Tesla vehicles, Apple CarPlay, and a growing set of developer products including Grok Voice, Voice Agent Builder, Grok Build, Grok Skills and Grok Imagine for image and video generation.
The Grok story is unusual among frontier labs for how tightly it is interwoven with Musk's other companies. Tesla ships Grok in its vehicles, where its own support page still describes the assistant as beta, and has been reported to be developing voice-intent support for its Full Self-Driving stack; SpaceX has offered discounted Starlink service to communities near contested xAI data centers, Musk says SpaceX engineers and staff from the coding startup Cursor have been reassigned to Grok development, and sworn Pentagon filings have described U.S. forces using Grok in connection with a large strike campaign. Its release process is also distinctive: roadmaps and capability claims often arrive first as Musk posts on X, ahead of or instead of model cards, benchmarks and documented access terms.
For practitioners, Grok is relevant on two levels. It is a rapidly iterating frontier model competing aggressively on price, with Grok 4.5 listed at $2 per million input tokens and $6 per million output tokens. It is also a case study in how safety, content-moderation and governance failures at a single lab, including recurring litigation over sexualized deepfake and CSAM-related content, a privacy regulator finding that Grok Imagine launched without an adequate privacy impact assessment, and a wrongful-termination suit from an engineer who raised safety concerns, translate into procurement and regulatory risk for the teams that adopt it.
The recent through-line for Grok is that its two most consequential storylines, military deployment and content-safety litigation, are both now documented in court filings rather than press releases. On August 4, coverage of a June 15 Justice Department filing established that xAI's Grok Gov Model was deployed through Maven Smart Systems during Operation Epic Fury; the attached Pentagon declaration says the system enabled U.S. forces to deploy more than 2,000 munitions against 2,000 targets within 96 hours. Crucially, the filing does not specify Grok's exact tasks or whether it selected targets or authorized weapons releases, and the earlier June reporting placed the same sworn statement from the Pentagon's chief digital and artificial intelligence officer, Cameron Stanley, inside a government brief defending xAI against an NAACP lawsuit over gas turbines at its Colossus 2 data center. For anyone assessing supplier governance, that is the practical shape of the disclosure: operational use is confirmed, the task boundary is not.
The legal picture moved just as fast, and it runs in both directions. X.AI's July 27 complaint challenges Minnesota's House File 1606, enacted as Chapter 72 in May, which prohibits operators of websites, applications and software from allowing users to nudify images of recognizable people and authorizes the attorney general to seek civil penalties of up to $500,000 for each unlawful access, download or use. On July 31, U.S. District Judge Donovan W. Frank denied the emergency request in a two-page order that pointed to the company filing on July 29, nearly three months after the law was signed and three days before it took effect; the statute took effect August 1, Minnesota's response is due August 12, X.AI's reply August 17, and a preliminary-injunction hearing is set for August 19 with the First Amendment challenge unresolved. Two weeks earlier xAI took the plaintiff's seat, filing a July 14 breach-of-contract complaint against Terry Wayne Harwood over allegations he used multiple Grok accounts to generate or alter sexualized images of adults and minors; the complaint claims 52,222 account suspensions and 73,604 reports to the National Center for Missing & Exploited Children in 2026, self-reported figures that have not been independently verified. The separate deepfake-CSAM class action was amended on July 7 to add two more anonymous plaintiffs. Against that backdrop the product cadence continued: SpaceXAI launched Grok 4.5 on July 8 with public availability advertised for July 9 and official docs listing $2 per million input tokens and $6 per million output tokens, landing in the same decision window as OpenAI's July 9 GPT-5.6 rollout. Musk had said on June 28 that the model was already in private beta at SpaceX and Tesla, built on a 1.5-trillion-parameter V9 foundation that finished training on May 26 and supplemented with data from the Cursor coding environment; no benchmarks or system card have been published, which leaves the Opus-class framing a vendor claim and makes cost per completed task, latency and structured-output reliability the only things a team can actually measure.
The nearest checkable date is the August 19 preliminary-injunction hearing in St. Paul on X.AI's challenge to Minnesota House File 1606, preceded by Minnesota's August 12 response and X.AI's August 17 reply; the July 31 order decided nothing on the merits, so whether the First Amendment argument survives is still open, as is whether other states with similar nudification statutes draw the same challenge. On the military thread, watch whether any filing or oversight process clarifies what the Grok Gov Model actually did inside Maven Smart Systems, since the June 15 Justice Department filing does not specify its tasks or whether it selected targets or authorized weapons releases. On the model itself, watch whether xAI publishes benchmarks or a system card for Grok 4.5 that would let the Opus-class claim be tested independently, and whether the $2 and $6 per million token pricing holds as usage scales. On safety governance, the open questions are whether the 52,222 account suspensions and 73,604 NCMEC reports cited in xAI's complaint against Terry Wayne Harwood are ever corroborated outside a filing, whether the amended deepfake-CSAM class action adds further plaintiffs, and what follows the Office of the Privacy Commissioner of Canada's finding that X Corp. and xAI violated federal private-sector privacy law by launching Grok Imagine without an adequate privacy impact assessment. On the corporate side, track whether the $60 billion Anysphere acquisition closes in the third quarter of 2026 as expected, whether the reserve SpaceX disclosed for potential legal costs moves in later filings, and whether Devin Kim's wrongful-termination suit reaches discovery.
No. SpaceXAI launched Grok 4.5 on July 8 with public availability advertised for July 9, but has not published benchmarks or a system card, so the comparison to Anthropic's Claude Opus remains a vendor claim traced to Musk's public statements. The measurable properties are the published prices, $2 per million input tokens and $6 per million output tokens, and whatever your own workload testing shows on latency, structured-output reliability and cost per completed task.
It establishes deployment, not decision authority. A June 15 Justice Department filing says the Grok Gov Model was deployed through Maven Smart Systems during Operation Epic Fury, and the attached Pentagon declaration says the system enabled U.S. forces to deploy more than 2,000 munitions against 2,000 targets within 96 hours. The filing does not specify Grok's exact tasks or whether it selected targets or authorized weapons releases. Reporting also notes Maven Smart Systems initially used Anthropic's Claude.
In Minnesota, yes, for now. The law took effect August 1 after Judge Donovan W. Frank denied X.AI's temporary restraining order on July 31 in a two-page order that turned on the company's nearly three-month delay in seeking relief, not on the merits. It places duties on service operators, not only end users, and authorizes civil penalties of up to $500,000 for each unlawful access, download or use. The constitutional challenge is unresolved, with a preliminary-injunction hearing set for August 19.
Treat them as litigation exhibits, not audited disclosures. The 52,222 account suspensions and 73,604 reports to the National Center for Missing & Exploited Children in 2026 appear in xAI's own July 14 breach-of-contract complaint against Terry Wayne Harwood and have not been independently verified. Separately, the Office of the Privacy Commissioner of Canada found that Grok Imagine launched without an adequate privacy impact assessment, that the assessment was not completed until March after the July 2025 launch, and that it did not accurately reflect security, safety and privacy risks.
Jurisdiction-specific image-generation rules are now enforceable law rather than proposals, and the vendor is an active party in multiple proceedings. Beyond Minnesota, the European Parliament voted 569 to 45 to ban AI systems generating sexualized deepfakes, New Zealand has advanced criminalization, British MP Jess Asato filed a High Court claim over Grok-generated sexualized images, and Apple privately warned xAI in January that it would pull the standalone Grok app unless the behavior stopped, approving a later remediation submission. Access controls, acceptable-use policy, advertising claims and abuse-reporting workflows are all in scope.
The evidence is mixed and some of it is coerced. The New York Times reported that Musk required banks, law firms, auditors and other advisers on SpaceX's IPO to purchase Grok subscriptions, with some banks agreeing to integrate it and potentially commit deals worth tens of millions of dollars. Interactive Brokers added Grok alongside ChatGPT and Claude connectors, with AI-generated order instructions requiring client review in a dedicated tab. On the other hand, Reuters found Grok in only 3 of more than 400 publicly identified federal AI-use records from 2025.
Latest coverage
Showing newest first with full-topic pagination
xAI's Grokipedia has not updated articles or processed suggested edits since April 24, according to an Aug. 5 Lawfare investigation. Lawfare examined 34,519 pages with suggested edits totaling 225,496 recommendations and found no accepted or rejected correction dated within the previous three months. The Verge reported that xAI did not immediately respond to a request for comment.
A June 15 Justice Department filing says xAI's Grok Gov Model was deployed through Maven Smart Systems during Operation Epic Fury. The attached Pentagon declaration says the system enabled U.S. forces to deploy more than 2,000 munitions against 2,000 targets within 96 hours, but the filing does not specify Grok's exact tasks or whether it selected targets or authorized weapons releases.
On July 31, U.S. District Judge Donovan W. Frank denied xAI's emergency request to block Minnesota's AI nudification law before it took effect August 1. The two-page order says xAI's nearly three-month delay suggested any harm was not immediate. The court will hear xAI's request for a preliminary injunction on August 19; the broader First Amendment challenge remains unresolved.
Minnesota's nudification law took effect on August 1 after U.S. District Judge Donovan Frank denied X.AI's request for an immediate temporary restraining order. The July 31 order did not decide whether the law is constitutional; it set an expedited schedule for the company's preliminary-injunction request while the challenge proceeds.
xAI filed a civil breach-of-contract lawsuit on July 14 against Terry Wayne Harwood, alleging that he used multiple Grok accounts to generate or alter sexualized images involving adults and minors in violation of xAI policies. The federal complaint seeks unspecified damages and a permanent ban from Grok; it also claims xAI suspended 52,222 accounts and sent 73,604 reports to the National Center for Missing & Exploited Children in 2026, figures that have not been independently verified.
A July lawsuit alleges that a man used Grok to create thousands of sexualized images of his stepdaughter and that xAI's safeguards and reporting response were inadequate. The widely cited 1.8 million figure in the frozen title comes from earlier monitoring summarized in congressional material; it is context about the broader Grok deepfake crisis, not confirmation of the facts alleged in this specific complaint. Other suits involving Baltimore, teenagers, and public figures are separate legal events and should not be treated as corroboration of one another. The immediate issue is whether platform controls, reporting processes, and evidence preservation met legal duties when a named victim and alleged misuse were identified.
Starlink's support page and multiple reports say SpaceX is applying a 50% monthly Starlink discount for eligible Memphis-area addresses near contested xAI data-center infrastructure. Business Insider, PCMag, Tom's Hardware, and Gizmodo connect the offer to local backlash over xAI or SpaceXAI data-center power, noise, and emissions issues, while Bloomberg is paywalled but on-topic. For AI infrastructure practitioners, the takeaway is that community acceptance, permitting, and environmental scrutiny can become operating risks for compute clusters; consumer incentives may help goodwill, but they do not resolve legal or regulatory exposure.
On July 8, 2026, Tesla was reported to be developing voice-intent support that would let drivers give contextual instructions to FSD, including which driveway, house, or local landmark should guide a trip endpoint. Teslarati and Benzinga tie the update to Tesla VP of AI Software Ashok Elluswamy's X confirmation, while Tesla's support page still describes Grok as beta and separates current navigation requests from vehicle-control commands. For autonomy teams, the important shift is not the assistant UI but the validation problem: free-form speech, remembered locations, and safety-critical planning create new edge cases around ambiguity, latency, privacy, and driver supervision.
SpaceXAI launched Grok 4.5 on July 8, 2026, with public availability advertised for July 9 and official docs listing $2 per 1M input tokens and $6 per 1M output tokens. The practical question is not whether the model is "Opus-class," a claim still needing independent benchmarks, but whether its coding, tool-use, finance, and legal-workflow positioning can lower cost per completed task. Cursor says the release is built for long-running tool use, while Mashable and PYMNTS tie the rollout to Elon Musk's public cost and efficiency claims. Teams should benchmark latency, structured-output reliability, and real workload cost before moving assistants or agents onto it.
OpenAI is preparing a July 9, 2026 public rollout of GPT-5.6 while Elon Musk said SpaceXAI would make Grok 4.5 public, putting two frontier-model launches into the same decision window for AI teams. Axios reported Commerce Department clearance for broader OpenAI access, Reuters-syndicated coverage said OpenAI delayed the release after a U.S. government request, and OpenAI's own materials describe Sol, Terra and Luna as a limited-preview family moving toward wider availability. For practitioners, the signal is operational: benchmark gains, cyber and bio risk ratings, token pricing, and government-access constraints all have to be evaluated before swapping models into production workflows.
Reuters and The Information reported on July 7, 2026 that SpaceXAI and Cursor could launch their first jointly developed AI model as soon as July 8, citing an internal memo. Reuters said it could not independently verify The Information's report, and Cursor declined to comment. For practitioners, the key issue is not the launch date alone: a SpaceXAI-Cursor model would test whether a coding-tool company with new compute access can reduce dependence on third-party frontier models while still serving developers who expect model choice.
On July 7, 2026, CyberScoop reported that the Grok deepfake-CSAM class action against xAI was amended to add two more anonymous plaintiffs, bringing new allegations about minors' real photos being used to generate abusive images. Reuters separately reported Baltimore's March suit over alleged Grok sexual deepfakes, and Law.com covered the earlier California class action. For practitioners, the operational lesson is narrow but serious: multimodal platforms need enforceable upload controls, output-retention policies, abuse evidence workflows, and law-enforcement response paths that can survive scrutiny when CSAM or nonconsensual imagery is alleged.
xAI added 21 new multilingual voices to Grok Voice on July 6, 2026, alongside naturalness upgrades to the original five voices, with availability across the realtime Voice Agent API, Text to Speech API, and Grok Voice Agent Builder. For voice-agent builders, this is less a consumer feature than new production infrastructure: a broader, natively multilingual voice roster lowers the friction of standing up localized, brand-specific assistants for support, education, commentary, and advertising workflows. Teams still need to separately evaluate transcription accuracy, latency, and escalation behavior before putting any of these voices in front of customers.
SpaceX's former xAI operation is being presented in new reporting as SpaceXAI, extending a February $1.25 trillion SpaceX-xAI combination into an orbital-AI infrastructure story. The strongest corroboration is older reporting from the Los Angeles Times and Satellite Today, plus an FCC public notice showing SpaceX sought authority for up to one million orbital data-center satellites. For practitioners, the useful takeaway is not to assume near-term orbital compute capacity; it is to track whether space-based AI moves from valuation narrative and filings into usable SLAs, power budgets, latency numbers, and compliance controls.
Tesla will cap employee AI spending at $200 per week starting July 6, according to an internal memo first reported by The Information and detailed further by Electrek, after software engineers were burning through "thousands of dollars' worth of tokens each week." Spending above the cap requires sign-off, but the policy excludes beta versions of Grok and other xAI products, an exemption that gives Musk's own AI company a built-in cost advantage even though Electrek reports Tesla engineers largely prefer Anthropic's Claude. The reversal comes just months after Tesla pushed staff to adopt AI more aggressively, echoing similar spending caps recently imposed at Uber, Meta, and Walmart.
xAI launched Voice Agent Builder in beta on July 1, 2026, a no-code platform that lets developers and businesses build production voice agents on Grok Voice in under two minutes. The tool bundles telephony, document retrieval, tool-calling, guardrails, and call observability into one speech-to-speech interface, replacing the fragmented three-vendor stack (speech-to-text, LLM, text-to-speech) most teams currently use. Pricing starts at $0.05 per minute of agent audio plus $0.01 per minute for telephony, undercutting established voice AI vendors like ElevenLabs and Vapi. According to xAI, its Grok Voice Think Fast 1.0 model scores 67.3% on the company's own tau-voice Bench, ahead of Gemini 3.1 Flash Live and GPT Realtime 1.5, though that benchmark is self-administered and not yet independently verified.
SpaceX showed investors and stakeholders a prototype AI hardware device, described as slimmer than an iPhone, ahead of its recent initial public offering, according to The Wall Street Journal. The handset-like device reportedly runs a proprietary operating system, uses xAI technology (the Grok-maker SpaceX acquired earlier in 2026), and is planned to use a Qualcomm Snapdragon chipset. SpaceX told investors the project is early-stage and could change or be cancelled, and Elon Musk has publicly disputed the report, calling it "utterly false." The story matters as another signal that non-traditional hardware players are chasing dedicated consumer AI devices, a category that has so far struggled commercially with products like the Humane AI Pin and Rabbit R1.
SpaceX is offering Memphis-area residents a 50% discount on Starlink internet service, cutting standard monthly plans (normally $55 to $135) to as low as $27.50, with no upfront hardware costs for new or existing customers. The offer, announced June 30, 2026 via posts from SpaceX VP of Starlink engineering Michael Nicolls and Elon Musk, is tied to xAI's Colossus data center campus in Memphis and nearby Southaven, Mississippi, which has faced lawsuits and community complaints over gas-turbine noise and emissions. For AI practitioners, the episode shows how compute buildouts increasingly pair local consumer incentives with infrastructure expansion, even though satellite last-mile service does not substitute for the fiber and peering large training campuses require.
For practitioners building with AI coding agents, the recurring tax is supervision: most agents stall at every step, waiting for a human to review output, approve a command, or restart a session that ended mid-task. xAI's new /goal mode in Grok Build targets exactly that overhead by letting the agent run a bounded objective to completion without a person in the loop. According to xAI's announcement on June 22, 2026, a developer issues a single objective, such as migrating an authentication module to a new API, and Grok Build plans an approach, decomposes it into a progress checklist, and executes each item while verifying its own work by reviewing code, inspecting web pages, or running scripts before marking the task complete. xAI added status, pause, resume, and clear controls so users can monitor or steer a long-running job without breaking its flow. The mode is available now in the Grok Build CLI. Reported coverage describes a two-model pipeline pairing a Composer planning model with Grok Build for execution and verification.
Elon Musk said on X that SpaceX has reassigned "a few dozen" senior Starlink and Starship engineers, plus staff from AI coding startup Cursor, to work on the Grok model family, according to Business Insider. Musk said Grok 4.5, built on a 1.5-trillion-parameter architecture, is now in private beta at Tesla and SpaceX, and that SpaceX plans to release new models "trained from scratch" every month this year. The move follows SpaceX's $60 billion agreement, announced June 16, 2026, to acquire Cursor's parent company Anysphere, expected to close in Q3 2026; SpaceX absorbed Grok developer xAI in February 2026 and completed a record $75 billion IPO on June 12, 2026. For AI/DS practitioners, the personnel and compute claims trace to Musk's own social posts, not an engineering report or independent benchmark.
For teams tracking the frontier-model race, the signal worth watching is not a public launch but who gets early access. Elon Musk said on June 28, 2026 that xAI's Grok 4.5 has entered private beta at SpaceX and Tesla, its first deployment ahead of a wider release. According to Musk, Grok 4.5 is built on xAI's 1.5 trillion-parameter V9 foundation model, which finished training on May 26, and adds data from the Cursor coding environment in supplemental training. Musk claimed early evaluations put its performance close to, and potentially above, Anthropic's Claude Opus. xAI has not published benchmarks or a system card, so the capability claims remain unverified. The choice to dogfood the model inside Musk's own engineering-heavy companies suggests a focus on technical reasoning and code over consumer chat.
xAI has made Grok Imagine Video 1.5 Preview available as a developer-facing video-generation model, according to the official xAI announcement page dated June 16, 2026 and xAI's accessible model documentation. The model is listed in xAI Docs as grok-imagine-video-1.5, part of the Grok Imagine family. For practitioners, the important shift is distribution: Grok's video generation is no longer only a consumer-facing feature inside the Grok product; it is being exposed as an API model that developers can evaluate for creative tools, marketing workflows, media agents, and product experiences. The practical questions are model quality, prompt control, latency, moderation, provenance, and review workflows before generated video is allowed into production.
A report in The Information, cited by Forbes, finds that Grok, the chatbot from xAI, receives most of its activity from pornographic images and videos, adult role-play chats and large volumes of erotic stories, according to Forbes reporting on the investigation. Forbes reports that Similarweb data shows a 22% drop in Grok web traffic between January and May. Tugatech reports that documents prepared for xAI's IPO disclose a reserve of about 490 million euros to cover potential legal costs. Forbes and other reporting also cite multiple complaints and at least one lawsuit alleging sexualized deepfakes and altered images of underage people associated with Grok usage.
Interactive Brokers expanded its agentic trading ecosystem by adding ChatGPT and Grok integrations, alongside its existing Claude connector, the company announced via press coverage this week (AFP; Finance Magnates). The integrations let clients link existing IBKR accounts through certified AI connector marketplaces without creating new accounts or sharing passwords or API keys with AI providers, and AI-generated order instructions for equities, ETFs, options, futures and futures options must be reviewed and approved by the client in a dedicated "AI Instructions" tab before submission (AFP; Finance Magnates; Binance summary). CEO Milan Galik was quoted describing growing investor interest in natural-language access to markets (AFP). Brokers embedding third-party LLM connectors lowers friction for natural-language workflows, but raises operational and oversight questions for trading teams and compliance functions.
In a legal filing reported by Reuters, The Independent, and other outlets, the Pentagon's chief digital and artificial intelligence officer, Cameron Stanley, said in a sworn statement that the xAI chatbot Grok was used to fire "2,000 munitions at 2,000 distinct targets within 96 hours" during Operation Epic Fury, the administration's name for the Iran strikes, per Common Dreams and Yahoo News. The statement appeared in a government brief defending xAI against an NAACP lawsuit alleging the company operated unpermitted gas turbines at its Colossus 2 data center, per reporting in New Republic and Economic Times. The filing also links Grok use to the military's Maven Smart Systems (MSS), which reporting says initially used Anthropic's Claude, per Economic Times/Reuters. This disclosure formalizes the operational use of a commercial generative AI model in kinetic targeting and raises urgent questions about auditability, legal liability, and supplier governance.
In a sworn Justice Department filing cited by multiple outlets, Cameron Stanley, the Pentagon's chief digital and artificial intelligence officer, wrote that Grok, the chatbot developed by xAI, was used to help fire more than 2,000 munitions at 2,000 distinct targets within 96 hours, according to reporting by The Independent and Yahoo. The filing framed continued operation of xAI data centers as "a matter of paramount national security," per Yahoo. Separate reporting in Yahoo said U.S. military investigators believe American forces using AI-driven targeting were likely responsible for a strike on a school in Minab that killed at least 175 people, mostly children. Earlier reporting has documented Grok producing antisemitic outputs and nonconsensual sexualized imagery, per DailyKos.
According to Reuters and Bloomberg, former xAI engineer Devin Kim filed a lawsuit in California state court alleging he was wrongfully fired in September 2025 for repeatedly raising safety concerns about the chatbot Grok. The complaint, which TechCrunch and Insurance Journal reviewed and cited, names xAI and parent SpaceX and accuses Kim's supervisor, xAI co-founder Jimmy Ba, of abruptly firing Kim shortly before a planned presentation on safety. The suit quotes Kim saying xAI's "failure to prioritize AI safety, particularly with respect to Grok, virtually guaranteed that the Company would commit unlawful acts," per Reuters. The nonprofit Center for AI Safety announced it had named Kim as president, as reported by Reuters and Insurance Journal. xAI and SpaceX did not immediately respond to requests for comment, Reuters reports.
A coalition called Safe AI Now (SAIN) staged a Times Square protest on June 11, 2026, deploying a giant inflatable caricature of Elon Musk with the slogan "SpaceX's Grok makes AI child porn," organizers told reporters (CNET; Business Insider; Futurism). The effigy stood outside Nasdaq and was displayed the day before SpaceX's high-profile IPO (Business Insider; CNET). SAIN said the inflatable was meant to highlight alleged harms from Grok, which activists link to sexualized deepfake images of children that circulated on social media; SAIN warned investors about potential legal and regulatory liabilities (CNET; Futurism). Business Insider and Futurism note SpaceX's S-1 filing acknowledged that Grok's NSFW mode could create "heightened risks" and "reputational harm." The protest group and media coverage say SpaceX did not immediately respond to requests for comment (CNET).
The Office of the Privacy Commissioner of Canada released a report finding that Grok Imagine, an AI image-generation tool, was launched without an adequate privacy impact assessment. The report, cited by Pixel Envy, states that a privacy impact assessment for Grok Imagine was not completed until March, after the model's July 2025 launch, and that the assessment "did not accurately reflect" risks to security, safety and privacy. The commissioner, Philippe Dufresne, found that X Corp. and xAI violated Canada's federal private-sector privacy law. The report says the lack of safeguards enabled users worldwide to create and share non-consensual, sexualized deepfakes, many targeting women and children.
A former xAI engineer, Devin Kim, filed a lawsuit in California state court alleging he was wrongfully terminated after raising safety concerns about the chatbot Grok. According to reporting by Reuters, Bloomberg, and TechCrunch, the complaint says Kim repeatedly warned supervisors that xAI was failing to prioritise safety and that Grok could foment discrimination and enable dissemination of weapons information. The suit names xAI and parent SpaceX and alleges Kim was fired in September 2025 shortly before a planned internal safety presentation, per Reuters and Bloomberg. TechCrunch reports the complaint cites episodes where Grok produced extremist and harmful outputs. The nonprofit Center for AI Safety last week named Kim its president, and xAI and SpaceX did not immediately respond to media requests, according to Reuters.
