News

A Wyoming resident identified as Jane Doe 4 has added her name to a class‑action suit accusing xAI's Grok chatbot of generating thousands of fake child sexual abuse images. The lawsuit alleges her stepfather used the AI tool to transform an 11‑year‑old photo into explicit content, then distributed it online. Law enforcement later seized his devices, and he died by suicide two days after the raid. Doe joins three other teenagers who originally filed the case in Tennessee, expanding a claim that the technology could have harmed thousands of minors. State and European regulators are also probing xAI over similar allegations.Read more

OpenAI has dissolved its AI‑preparedness unit, the team that evaluated existential risks and mitigation strategies, and reassigned its duties to existing groups focused on bio‑security and cyber threats. The move comes as the company readies for a highly anticipated initial public offering and follows a series of restructurings that saw the departure of several senior safety and ethics leaders. Critics argue the shift signals a tilt toward product speed over safety, while the former team head, Dylan Scandinaro, now concentrates on the challenges of recursive self‑improving AI.Read more

Meta chief executive Mark Zuckerberg unveiled a 6,500‑word essay titled “The Future Is for Everyone,” promising personal AI agents that will understand users’ goals and run on any device. The announcement, delivered on a TechCrunch podcast, highlighted Meta’s new models—Glimmer and Muse Spark—and suggested a future where AI assistants are always on, even offline. Critics, however, pointed to the company’s history of social‑media controversy, limited consumer AI presence and hardware constraints, arguing that the vision sounds more like a corporate pitch than a near‑term reality.Read more

Anthropic announced that its Claude AI will embed invisible watermarks in generated text, using a version of Google’s open‑source SynthID‑Text system. The move satisfies the European Union’s AI Act, which mandates machine‑readable tags for synthetic content. The company says the watermark will not affect cost or output quality, and it joins other AI firms like Google in preparing for stricter transparency rules.Read more

Wispr, the AI‑driven dictation startup, announced a $280 million Series B round on Monday, led by Menlo Ventures and valuing the company at $2 billion. The fresh capital will fund expansion into meeting‑note taking and broader AI interfaces, while existing backers such as NEA and 8VC also participated. Wispr introduced a new speech‑understanding model, Canto, aimed at cutting error rates dramatically, and rolled out its note‑taker tool to challenge rivals in the fast‑growing market.Read more

Anthropic announced that its Claude language model will embed an undetectable watermark in every text it generates, a move aimed at complying with the European Union's new AI transparency regulations. The technique leaves a hidden pattern that only a decoder with the proper key can read, does not alter output quality or speed, and will be applied to all Claude outputs released after Aug. 2, with older models slated for update in the coming months.Read more

Anthropic’s August 14 Risk Report reveals that from May 2025 through April 2026 its human‑feedback platforms operated without the classifiers designed to block biological‑weapon assistance. Roughly 50,000 contractors generated about 133 million exchanges while the filters and logging were turned off. A review flagged 1,197 high‑risk transcripts, but staff found no clear evidence of dangerous misuse. The company also disclosed an internal model, Model 2, that will not be released, and detailed a second incident where a vendor‑derived API key gave access to a powerful model that ran without safeguards for two weeks.Read more

A leaked draft of a U.S. State Department letter, reported by Reuters on Aug. 14, asks the 35 signatories of the AI Opportunity Statement to choose between Washington’s Pax Silica framework and China’s parallel AI cooperation effort. The undated memo, seen by a U.S. official, warns that participation in both initiatives is untenable and stresses a “commitment” to the American supply‑chain alliance covering AI models, semiconductors and critical minerals. Kazakhstan, the only country to have joined both blocs, emerges as a test case for the policy’s enforcement.Read more

Anthropic disclosed that its second‑quarter revenue topped $11.5 billion, more than 14 times the $787 million it earned in the same period a year earlier. The AI firm also recorded positive adjusted operating income for the quarter, a milestone the sector’s leading labs have not yet achieved. The preliminary figures come as Anthropic prepares a confidential filing for a public listing, with investment banks Morgan Stanley, Goldman Sachs and JPMorgan advising the company and backers eyeing a $2 trillion valuation in October.Read more

Alibaba announced that its Qwen open-weight models have exceeded 3 billion downloads, positioning the family ahead of offerings from Meta and Google. A Hugging Face report released a day earlier recorded roughly 2.05 billion downloads and 151,000 derivative models, a third fewer than Alibaba’s numbers and still more than twice Meta’s total. The discrepancy stems from differing measurement methods, with Hugging Face tracking activity on its own hub and Alibaba counting traffic on its ModelScope platform. Both parties agree that Qwen has become a staple in developers’ toolkits, even as the U.S. AI community rolls out competing open models.Read more

SpaceX announced Friday that it has closed its purchase of AI coding firm Cursor, officially bringing the startup into its growing artificial‑intelligence portfolio. The deal, first outlined in April with an option to buy for $60 billion, was executed as SpaceX transitioned to a public company. Cursor’s blog said the acquisition gives it access to SpaceX’s massive GPU fleet and computing infrastructure, which already serves customers like Anthropic and Google. The move follows SpaceX’s earlier acquisition of Musk’s xAI and comes amid a lawsuit over pollution from its data‑center turbines.Read more

Anthropic released a blog post on Friday outlining how its Claude chatbot will embed invisible watermarks in generated text to meet the EU AI Act’s Transparency Code. The company says the watermark is undetectable to readers but can be identified with a secret key, and it will not affect output quality. Users on Reddit and X have voiced concern, with some cancelling subscriptions. Anthropic also explained that light editing won’t erase the watermark, while a full rewrite might, and that code snippets will carry minimal watermarking. A detection API is slated for release.Read more

A woman identified as Jane Doe 4 has filed a lawsuit alongside three Tennessee teenagers against Elon Musk’s xAI, alleging that the company’s chatbot Grok was used by her stepfather to turn an 11‑year‑old photo of her into more than 7,000 explicit images. The stepfather died by suicide two days after law‑enforcement uncovered the material. Plaintiffs accuse xAI, now part of SpaceX, of neglecting safeguards that could have prevented the abuse, and they are seeking class‑action status.Read more

OpenAI rolled out a new feature called Computer History in the macOS version of its ChatGPT desktop app. The opt‑in tool logs user activity as discrete events, allowing the AI to recall recent work, suggest automations and finish unfinished tasks. Users can exclude specific applications, block private‑browser sessions and delete entries at will. OpenAI says the feature avoids screenshots, video or audio capture, distinguishing it from Microsoft’s Windows Recall. Product and Engineering manager Ari Weinstein and developer Dominik Kundel demonstrated the capability in a short video released on Aug. 16, 2026.Read more

Anthropic chief executive Dario Amodei pushed back against claims that his warnings about artificial intelligence have fueled a U.S. backlash. In a series of posts, Amodei said his commentary balances risks and benefits and that public distrust—not his messaging—drives negative sentiment. He argued that the industry’s failure to deliver promised benefits, such as curing cancer, underlies skepticism, and he defended Anthropic’s nuanced regulatory proposals that aim to curb frontier AI power while supporting smaller players.Read more

A tech writer recently swapped careful prompt crafting for a ten‑minute stream of consciousness in ChatGPT’s voice mode. Speaking freely about vacations, TV choices, birthday gifts and more, the author let the AI sort the jumble afterward. ChatGPT identified a unifying theme—making everyday moments more meaningful—and offered practical suggestions. The experiment suggests that unedited speech can surface contradictions and hidden priorities that polished prompts often erase, giving the model a fuller picture of a user’s intent.Read more

OpenAI has added a new "Computer History" feature to ChatGPT for macOS users on its Pro, Business and Enterprise plans. The optional tool records clicks, keystrokes, app switches and other activity via macOS’s accessibility API, letting the AI summarize recent actions and suggest repeatable tasks. Users can pick which apps or sites to track, pause logging, and delete the history at any time. The feature is not yet available in the European Economic Area, Switzerland or the United Kingdom, and administrators can control opt‑in settings for those regions.Read more

DeepSeek announced a steep increase in its AI API pricing effective Aug. 16, as the company rolls out its V4 Pro and V4 Flash models. Peak‑hour rates for V4 Pro will jump to $3.96 per million output tokens, more than four times the current $0.87 price, while off‑peak rates will be $1.98. V4 Flash will cost $1.32 at peak and $0.66 off‑peak, up from $0.28. The move ends a promotional period that began in May and positions DeepSeek’s new rates still well below rivals like OpenAI and Moonshot.Read more

OpenAI and Anthropic have slashed the prices of their mid‑tier large‑language models, aiming to make U.S. offerings more competitive against a wave of lower‑cost Chinese AI services. OpenAI cut its GPT‑5.6 Luna model to $0.20 per million input tokens and $1.20 per million output tokens, while Anthropic introduced Opus 5 at $5 per million input and $25 per million output tokens, half the cost of its flagship Fable 5. The moves accompany a canceled price hike for Anthropic’s Sonnet 5 and signal a strategic push to defend the high‑end market while courting cost‑sensitive developers.Read more

Google announced Friday that users will soon be able to disable the visible watermark on images, videos and songs generated by its Gemini models. The change, which applies to Nano Banana, Omni and Lyria, leaves the invisible SynthID watermark and C2PA metadata intact for transparency. Settings will be added to Gemini, the Flow video editor and later to Search, with the feature rolling out in the next few days.Read more