#anthropic

Google, Anthropic, and OpenAI Unveil Cyber AI Models, Safeguards, and Access Programs Google announced on Wednesday the release of Gemini 3.8 Flash Cyber, its most advanced cybersecurity model to date, and introduced the Fairwind Program to provide early access to trusted defenders. The initiative targets organizations such as governments, healthcare providers, and telecommunications services, offering them advanced tools to build defenses against emerging threats. Google emphasized that the program grants these entities a strategic advantage by enabling them to prepare for cyber risks before they materialize. The company is currently collaborating with over 650 global partners, including CrowdStrike, Datadog, Palo Alto Networks, and Snowflake, to expand the program’s reach. Gemini 3.8 Flash Cyber represents a significant upgrade from its predecessor, Gemini 3.5 Flash Cyber, which was unveiled a month earlier. The latest model demonstrates frontier-level performance in autonomous vulnerability discovery, surpassing larger models from competitors like Anthropic’s Mythos 5 and OpenAI’s GPT-5.6 Sol and GPT-5.5-Cyber. Google’s focus on equipping defenders with expertise over offensive capabilities was highlighted by Tulsee Doshi and Raluca Ada Popa, who noted the company’s prioritization of vulnerability fixing from the outset. Anthropic simultaneously launched Claude Fable 5.1 and Claude Mythos 5.1, each with distinct safeguard levels. Mythos 5.1, available only through trusted access programs, is designed for cybersecurity and life sciences applications. While Fable 5.1 is now used for identifying software vulnerabilities, Anthropic plans to retain Opus models for tasks like penetration testing and exploit generation. The company evaluated Mythos 5.#google #anthropic #openai #crowdstrike #datadog

OpenAI Faces Turbulent Reckoning as AI Race Intensifies Protesters gathered outside OpenAI’s headquarters in early August, holding signs demanding an end to the “AI race” and leaving chalk messages on the sidewalk. Inside, executives from the company’s most prominent clients previewed Astra, OpenAI’s next-generation AI models, during a demonstration that showcased the system’s ability to solve complex math problems and navigate desktop software with unprecedented speed. CEO Sam Altman, who had recently returned from a closed-door meeting with U.S. officials, emphasized the potential of Astra to act as a “persistent agent”—a virtual colleague capable of sustained, high-level task execution. He also predicted the model could enable users to “invent new things in a way that matters,” a claim he framed as a step toward artificial general intelligence (AGI). Yet the company’s recent trajectory has been marked by setbacks. Over the past year, OpenAI has lost its lead in the AI race to Anthropic, a rival founded by former OpenAI employees. Anthropic’s Claude Code, a coding-focused AI tool, has become a market leader, surpassing OpenAI in revenue and private valuation. Anthropic is now expected to be the first of the two companies to go public, with an IPO potentially as early as September. Meanwhile, OpenAI has faced a series of internal and external challenges, including leadership departures, legal battles, and a growing erosion of public trust. The company’s leadership team has undergone significant changes. Fidji Simo, the former Instacart CEO hired as Altman’s deputy, left after just a few months. Key figures from OpenAI’s safety, ethics, and research teams also departed, while Denise Dresser, the chief revenue officer, and Brad Lightcap, the former chief operating officer, resigned within eight months.#anthropic #openai #sam_altman #fidji_simo #hugging_face

Gavin Baker: Data Centers Are 'Awesome for Blue-Collar Americans' Gavin Baker, a venture capitalist and managing partner at Atreides Management, has become a vocal advocate for data centers, calling them a transformative force for working-class Americans. During a recent appearance on "The A16z Show," Baker emphasized that data centers are "really, really, really good" and have revitalized struggling small towns across the U.S. He argued that the rise of AI-driven data centers has created new opportunities for blue-collar workers, making trades like electrician, plumber, and HVAC technician more lucrative than traditional college degrees. Baker claimed that a college education is now "significantly NPV negative" due to the high costs and diminishing returns, while skilled tradespeople are earning "ungodly amounts of money" amid the AI boom. Baker highlighted the shift in labor markets, noting that many young people are opting out of undergraduate programs to pursue careers in data center operations. He cited an example of a data center electrician who skipped college and now earns $40 per hour, a stark contrast to the financial burden of higher education. This trend, Baker argued, reflects a broader anti-college sentiment in tech circles, where practical skills are increasingly valued over academic credentials. He also addressed environmental concerns surrounding data centers, acknowledging their high energy and water usage but asserting that the industry is "getting much better" at mitigating these issues through innovation and efficiency. The venture capitalist also pointed to geopolitical tensions, blaming China for an "organized, CCP-funded campaign" against data centers.#space_x #anthropic #atreides_management #gavin_baker #the_a16z_show

Anthropic Accelerates Fable 5.1 Launch Amid OpenAI's Astra Rivalry The AI industry is intensifying as Anthropic reportedly plans to release its next-generation flagship model, Fable 5.1, as early as tomorrow, positioning itself ahead of OpenAI’s upcoming Astra model. This strategic move aims to secure first-mover advantage in the competitive agent race, according to insiders and analysts. Anthropic’s decision marks a shift from its initial plan to wait for OpenAI’s Astra before launching Fable 5.1, as revealed by insider Leo, who previously predicted OpenAI would unveil Astra within weeks. He now asserts that Anthropic will not delay its release, citing the urgency to counter OpenAI’s advancements. Developers have shared screenshots of grayscale testing on Fable 5.1, with codenames “melon” and “marshmallow” appearing briefly in the Claude model list before being removed. These are believed to be early-access test versions of Fable 5.1 and Opus 5.1, respectively. AI leaker @Lentils80 noted that “melon” has reached a Fable-level checkpoint, while “marshmallow” is nearly as capable as a weakened Fable or a highly advanced Opus. However, both models were swiftly taken offline by Anthropic, following a pattern of rapid takedown after exposure, which aligns with the company’s historical release process for Claude products. The timeline of events reveals key milestones: on August 15, Anthropic’s Risk Report Appendix 6.6 mentioned a mysterious “Model 2,” described as “somewhat stronger than Mythos 5” with significant improvements in internal tasks. By August 20, Leo reported OpenAI’s Astra was expected within weeks, but Anthropic’s strategy shifted. On August 24, the codenames “melon” and “marshmallow” surfaced, only to be removed hours later. Grayscale users now anticipate an official release of Fable 5.1 and Opus 5.#anthropic #openai #fable_51 #astra #claire_code_team

Anthropic Unveils Claude Fable 5.1 and Claude Mythos 5.1: Advanced AI Models for Coding and Scientific Research Anthropic has launched two new iterations of its AI models, Claude Fable 5.1 and Claude Mythos 5.1, positioning them as the most advanced tools for coding, knowledge work, and scientific research. The models are functionally identical but differ in their safeguard configurations, with Fable 5.1 available to the general public and Mythos 5.1 reserved for trusted access programs focused on high-risk domains like cybersecurity and life sciences. The updates reflect Anthropic’s efforts to address user feedback on cost, data privacy, and safety measures. Fable 5.1 introduces significant cost reductions, with pricing estimated to be 25% lower than its predecessor for typical workloads, particularly when billed by token usage. This reduction stems from optimizations in cache read efficiency, which can lead to savings of up to 45% for highly agentic tasks. The model also features Enterprise Frontier Safeguards (EFS), a new system allowing enterprise customers to store data in cloud infrastructure they fully control, ensuring zero data retention while maintaining robust security. EFS will roll out in phases starting later this fall, with eligible customers able to use Fable 5.1 with zero data retention until the system is available. Safeguard improvements focus on reducing false positives, particularly in cybersecurity. Fable 5.1’s new safeguards block 60% fewer false positives than previous versions, partly due to its ability to identify software vulnerabilities without developing exploits. In biology, Anthropic has partnered with the U.S. government to create an access program for Mythos 5.1’s advanced capabilities, with enrollment expected to open soon for scientists.#nasa #anthropic #claude_fable_51 #claude_mythos_51 #millennium_investment_firm

Binance Now Lets AI Agents Trade, But Keeping Them in Check Is Largely Up to Users Binance, the world’s largest cryptocurrency exchange with over 300 million registered users, has launched a new platform called Agent OS that enables AI agents to analyze financial markets and execute trades on behalf of users. This development marks a significant step in integrating artificial intelligence into the management of real-world financial assets, as the platform allows developers to connect AI applications to Binance’s infrastructure. The initiative brings together existing tools such as Binance APIs, the Wallet Agentic Hub, x402 transaction verification, and the Skill Hub, alongside newly introduced support for the Model Context Protocol (MCP). The platform also integrates with popular AI tools like OpenAI’s ChatGPT and Codex, Anthropic’s Claude Code, and Cursor, enabling users to authorize agents to access market data, view account details, and perform trades. However, Binance has emphasized that users retain primary responsibility for ensuring the safety and ethical use of these AI agents. While the platform provides granular access controls, the exchange has not imposed strict limits on how much an AI agent can trade or lose. Instead, users are required to configure permissions and set boundaries through subaccounts, which act as isolated environments for agent activity. Jeff Li, Binance’s vice president of product, explained that the platform prioritizes user control, stating, “We put the power in users’ hands to give them the granular access control of what they can do through the agent.” Subaccounts are designed to restrict withdrawals by default, creating a sandboxed environment where agents can operate without direct access to users’ main funds.#anthropic #openai #binance #agent_os #jeff_li

Claude Will Put Invisible Watermarks On AI Text And Images—And The Internet Isn’t Happy Anthropic, the company behind the Claude AI model, has announced a new policy to embed invisible watermarks into text and images generated by its AI systems. The move, which was revealed on Tuesday, aims to comply with European Union regulations requiring transparency about AI-generated content. Under the policy, all text and media produced by Claude will carry embedded watermarks that are designed to persist even when the content is copied, pasted, or shared elsewhere. Additionally, files created by the AI, including images, will include digitally signed metadata to indicate their origin. However, the implementation details remain unclear, with Anthropic stating it will provide further guidance on how to identify the watermarks. The company also warned that the watermarks may not be foolproof, as they could be erased if text undergoes heavy editing or if a file’s metadata is lost due to format changes, screenshots, or other factors. The policy, which users cannot opt out of, will apply to all Claude models and products globally, including specialized tools like Claude Code and Claude Cowork. The announcement has sparked significant backlash from users on platforms like X (formerly Twitter) and Reddit, where many have criticized the move as intrusive and unnecessary. Some users argue that the watermarking system undermines the value of their work, particularly those who use Claude for tasks like proofreading or coding. For example, radio show host and blogger Erick Erickson expressed frustration on X, stating that he had switched from Grammarly to Claude for its superior proofreading capabilities but now faces the risk of his work being marked as AI-generated.#anthropic #claude_ai #openai #eu_regulations #erick_erickson

Anthropic Adds Machine-Readable Watermarks to Claude Text Under EU AI Act Rules Anthropic will begin embedding imperceptible machine-readable watermarks into text generated by its new Claude models starting August 2, 2026. This move aligns with transparency requirements under Article 50 of the EU AI Act, which mandates greater accountability for AI systems. The company has also signed the European Union’s Code of Practice for AI-generated content, reinforcing its commitment to compliance. The new watermarking system will apply globally, not just to European users, and will be integrated into all outputs from Claude models launched after August 2, 2026. The watermarks will be applied at the model level, ensuring they persist across all Claude products and platforms. For text, the system will use invisible signals that remain embedded even after users copy and paste responses. Some marks may survive editing, though Anthropic acknowledges that heavy modifications could reduce their detectability. The watermarking process will extend to supported files, such as images, through a separate system based on the Coalition for Content Provenance and Authenticity (C2PA) standard. Files will carry signed metadata to verify their origin and detect unauthorized alterations. Anthropic’s approach aims to provide platforms and organizations with a technical signal to identify AI-generated content. The company plans to publish detailed technical documentation to help users and third parties verify the presence of watermarks. This includes tools for detecting the marks, which could aid in tracking AI-generated material across workflows. However, Anthropic cautions that neither text watermarks nor file metadata should be considered definitive proof of authorship.#anthropic #claire #eu_ai_act #c2pa #code_of_practice_ai_generated_content

Anthropic Announces AI Text Watermarking Compliance with EU Regulations Anthropic, a leading AI model developer, has announced it will implement watermarking for text generated by its AI systems, including the Claude series, to align with European Union regulations. The company confirmed this policy update in a revised support page, emphasizing its commitment to adhering to the EU AI Act’s Transparency Code, which became enforceable on August 2, 2026. This regulation mandates that AI-generated or edited content be marked in a way that allows other systems to identify its origin. Under the new policy, all AI models released after August 2 will automatically include technology to watermark both textual and file-based outputs. For files, Anthropic will utilize the C2PA (Content Credentials for the Public Web) open standard, a framework designed to verify the authenticity and origin of digital content. The company also stated it will provide support for older models, ensuring the watermark remains intact even when users copy and paste the text. According to the support page, the watermark is embedded as part of the text itself, meaning it will persist through copying, pasting, and some editing processes. The watermark will be applied at the model level, ensuring its presence regardless of which Claude product or platform the text originates from. The implementation of watermarking extends to various Anthropic products, including the Claude platform API, Claude, Claude Code, Claude Cowork, and Claude Tag. This measure aims to enhance transparency and accountability in AI-generated content, particularly in contexts where distinguishing human-created text from AI-generated material is critical.#anthropic #claude #eu_regulations #eu_ai_act #c2pa

Anthropic Introduces Invisible Watermarks to Combat AI-Generated Text Misuse Anthropic, the artificial intelligence research lab behind the Claude series, has launched a new feature that embeds imperceptible watermarks into AI-generated text. The tool, designed to trace the origin of content created by Claude models, marks a significant shift in the company’s approach to transparency and accountability in AI output. The move aligns with Anthropic’s obligations under the European Union’s AI Act, which mandates greater oversight of AI systems, but the technology will apply to text generated globally, including through cloud providers. The watermark, described as “imperceptible,” is integrated directly into the text without altering its meaning or readability. It travels with the content when copied or pasted and may persist through some editing, though Anthropic acknowledges that heavy editing, paraphrasing, translation, or blending Claude’s output with other writing could potentially erase or obscure the mark. The company emphasized that the presence of a watermark does not definitively prove AI authorship, as even human editing or translation using AI tools might leave traces. This development comes amid growing concerns about the misuse of AI in creative and academic contexts. The publishing industry, in particular, has faced challenges from allegations of AI-generated content infiltrating original works. Earlier this year, Hachette Group, a major publisher, retracted Mia Ballard’s horror novel Shy Girl after claims surfaced that AI had been used to write parts of the book. Ballard denied using AI, stating that a freelance editor had unknowingly introduced AI-generated material into her work.#anthropic #claude #hachette_group #ai_act #aigenerated_text

Anthropic Pays AI’s Biggest Salaries. Its CEO Just Discovered People Might Take Them for the Money The AI industry is grappling with a growing crisis as elite researchers continue to shift between top labs, despite the astronomical salaries offered by companies like Anthropic. This week, the sector’s tensions reached a new level when Anthropic’s CEO, Dario Amodei, reportedly expressed concern that new hires are joining his lab primarily for financial incentives rather than a shared mission. The revelation sparked widespread online mockery, with many dismissing the issue as a trivial complaint from a company that pays some of the highest salaries in the field. However, beneath the humor lies a serious challenge: retaining top talent in an industry where financial compensation alone is no longer sufficient to secure long-term loyalty. The issue gained traction after a line from an Axios report about Anthropic’s talent wars went viral. The report highlighted Amodei’s worry that researchers are prioritizing pay over purpose, a sentiment that quickly became a punchline on social media. Engineers and industry insiders pointed out that Anthropic’s compensation packages are among the most generous in the AI world, often surpassing those of competitors like OpenAI. This irony—fearing that researchers are motivated by money while offering the highest salaries—has sparked debates about the sustainability of relying on financial incentives to attract and retain talent. The churn in the AI workforce is not a new phenomenon, but recent events have intensified the problem. Lilian Weng, a co-founder of Mira Murati’s former Thinking Machines Lab, became a focal point after she left her position last week and reappeared at OpenAI.#dario_amodei #anthropic #openai #lilian_weng #thinking_machines_lab

China's AI Models Challenge U.S. Tech Dominance with Open-Source Releases The global AI landscape has intensified as Chinese companies Alibaba, DeepSeek, and Moonshot.AI launch high-performance, open-source models that directly challenge U.S. rivals like Anthropic and OpenAI. The competition reached a critical point on Monday with the simultaneous release of Alibaba’s Qwen 3.8-Max and DeepSeek’s V4-Flash, models that are redefining cost-efficiency, scalability, and technical capabilities in the global AI market. Alibaba’s Qwen 3.8-Max, a 2.4 trillion-parameter model, marks the company’s first major open-weight release, positioning it as a direct competitor to OpenAI’s GPT-5.6 Luna and Anthropic’s Claude Sonnet 5. The model’s launch comes amid growing concerns from U.S. developers about the safety and origins of Chinese AI systems. Anthropic CEO Dario Amodei has publicly criticized open models from China, arguing they lack rigorous safety metrics and could pose risks if released without oversight. However, critics argue that proprietary models, which are tightly controlled by companies like Anthropic, are inherently more vulnerable to security threats due to their closed nature. DeepSeek’s V4-Flash model further intensifies the competition by offering a 284-billion-parameter solution that outperforms its 1.6 trillion-parameter predecessor, the V4 Pro, by nearly 14 percent on Artificial Analysis’ Intelligence leaderboard. At just 142 GB of GPU memory, the V4-Flash is designed for enterprise scalability, running efficiently on modest hardware. Its pricing strategy—$0.14 per million input tokens, $0.0028 per million cached tokens, and $0.28 per million output tokens—significantly undercuts OpenAI’s GPT 5.6 Luna, which costs $0.20 per million input tokens, $0.02 per million cached input, $0.25 per million cached write, and $1.#anthropic #openai #alibaba #deepseek #moonshot_ai

Moonshot AI's Kimi K3 Challenges U.S. AI Dominance with Cost-Effective Performance On Thursday, Moonshot AI, a Beijing-based startup, launched its latest large language model, Kimi K3, which claims to rival Anthropic’s Fable 5 in performance while costing significantly less. The model, described as the largest open-weight model ever released, reportedly outperformed Anthropic’s Opus 4.8 and OpenAI’s GPT 5.6 Sol in benchmark tests. Moonshot’s publicly shared evaluations placed K3 among the top three AI models globally, with one independent test from Arena.AI ranking it as the best available, surpassing Anthropic’s offerings. Industry experts initially underestimated the pace at which Chinese AI labs could close the gap with U.S. counterparts. Anthropic CEO Dario Amodei noted that a Chinese company achieving performance close to Fable 5 was expected to take six months, while Elon Musk had predicted such a breakthrough by early 2025. K3’s release, however, accelerated this timeline, highlighting the rapid progress of Chinese developers. Paul Triolo of DGA-Albright Stonebridge Group remarked that the Chinese AI ecosystem is “much better than people thought,” underscoring the shift in global AI dynamics. The model’s pricing strategy further disrupted the market. Kimi K3 is available for $15 per million output tokens, a stark contrast to Fable 5’s $50 for the same level of output. This cost advantage has raised concerns among investors who previously believed U.S. firms could maintain their lead through superior computing power spending. The K3 launch intensified an ongoing semiconductor selloff, with Taiwan Semiconductor Manufacturing Company dropping 7% despite strong quarterly profits. SoftBank, often linked to OpenAI, fell 9%, while Z.ai, a Chinese competitor, plummeted nearly 30% in Hong Kong. U.S.#dario_amodei #elon_musk #anthropic #moonshot_ai #kimik3

Wealthy AI Workers Drive San Francisco Housing Market Surge In the affluent Duboce Triangle neighborhood of San Francisco, a three-bedroom apartment in a historic Edwardian-era home has sparked significant attention, with its top half listed for nearly $3 million. The property, which has been opulently renovated, is unique in its payment terms: the seller is open to accepting shares in artificial intelligence companies like OpenAI or Anthropic instead of cash. This unconventional approach has drawn interest from tech professionals, including a young OpenAI employee who recently viewed the unit with his partner. The worker, who relocated to San Francisco two years ago for a technical role, currently rents and is considering negotiating with his employer to facilitate a stock-based transaction. The broader San Francisco housing market has seen unprecedented growth, with median home prices reaching a record high of $1.76 million as of May 2026. This surge has outpaced national trends, where U.S. home prices rose by just 1.4% in March and 2% in April and May. The city reclaimed its title as the most expensive U.S. market for homebuyers in March 2026, surpassing San Jose, and the trend has continued, with annual price increases of 19%, 14.5%, and 14.1% in March, April, and May respectively. Real estate economist Daryl Fairweather of Redfin attributes this explosion to the influx of wealth from the AI industry, noting that the Bay Area’s luxury zip codes—particularly Duboce Triangle—have experienced steep price jumps since OpenAI launched ChatGPT in late 2022. The AI boom has also revitalized San Francisco’s real estate landscape, halting the downturn caused by the pandemic. High salaries and signing bonuses for top AI talent, coupled with generous stock options, have created a surge in purchasing power.#san_francisco #redfin #anthropic #openai #duboce_triangle

Amazon’s $2 Trillion Empire Faces a Critical Turning Point Amazon’s market valuation has surged to $2.6 trillion as of July 2, 2026, driven by rapid growth in its cloud computing and artificial intelligence divisions. The company’s Q1 2026 earnings report revealed significant momentum, with revenue reaching $181.52 billion, a 16.61% year-over-year increase. Earnings per share (EPS) came in at $2.78, surpassing estimates by 68.18%, marking the fifth consecutive quarter of EPS beats. Net income of $30.25 billion included $16.8 billion in pre-tax gains from Anthropic holdings, a non-recurring item, while operating income rose 29.6% to $23.85 billion, reflecting a 13.1% corporate operating margin. The growth story is anchored by Amazon Web Services (AWS), which reported $37.59 billion in cloud revenue for Q1, a 28% year-over-year increase—the fastest pace in 15 quarters. AWS’s operating margin stood at 37.7%, highlighting its profitability. The company’s chip business, including Graviton, Trainium, and Nitro processors, achieved a $20 billion annual run rate with triple-digit growth. Advertising services revenue hit $17.24 billion, up 24% year-over-year, and now operates at a trailing rate above $70 billion. Unit growth in physical stores reached 15%, the highest since the end of the COVID-19 lockdowns. Amazon’s AI initiatives are also driving value. Amazon Bedrock processed more tokens in Q1 than all prior years combined, with customer spend on the platform growing 170% quarter-over-quarter. The company has secured significant AI infrastructure contracts, including 2 gigawatts of Trainium capacity for OpenAI through 2027 and up to 5 gigawatts for Anthropic. Meta is also listed as a customer.#amazon #aws #anthropic #openai #andy_jassy

Anthropic Launches Indian Rupee Billing for Claude AI Subscriptions Anthropic has expanded its global reach by introducing Indian rupee (INR) billing for its Claude AI subscription plans, enabling users in India to pay directly in local currency rather than U.S. dollars. This change applies to the Claude Pro, Claude Max, and Claude Team tiers, aiming to simplify payment processes, eliminate foreign exchange fees, and enhance accessibility for individual users, developers, and businesses in India’s rapidly growing AI market. The free version of Claude remains available, while paid subscriptions offer higher usage limits, priority access to new AI models, and advanced features such as Claude Code and premium reasoning capabilities. The shift to INR billing means Indian users will see subscription prices displayed in rupees, with Goods and Services Tax (GST) included in the listed amount. This eliminates the need for separate tax calculations during checkout, providing a more transparent and predictable payment experience. Previously, Indian customers had to pay in U.S. dollars, which often led to additional costs such as foreign exchange conversion charges, international transaction fees, currency fluctuations, and banking expenses. By adopting local currency billing, Anthropic seeks to reduce these hidden costs and improve user convenience. The decision reflects India’s significance as a key market for generative AI, driven by factors such as a large and active developer community, rapid adoption of enterprise AI solutions, an expanding startup ecosystem, and growing demand for coding assistants. The move also aligns with Anthropic’s broader strategy to strengthen its international presence, as the country continues to emerge as a hub for AI innovation.#india #indian_rupee #anthropic #goods_and_services_tax #claire_ai

OpenAI Proposes U.S. Government Own 5% Stake to Address Political Blowback OpenAI has proposed granting the U.S. government a 5% ownership stake in the company, according to a report by the Financial Times. The potential stake, valued at approximately $42.6 billion, would be based on the artificial intelligence startup’s recent $852 billion valuation following a record-breaking funding round in March. OpenAI CEO Sam Altman reportedly argued that this move would allow the public to share in the financial benefits of AI advancements. The proposal aims to mitigate growing political pressure in Washington, where lawmakers and regulators have increasingly scrutinized the influence and potential risks of large tech firms. The plan, outlined in discussions with the Trump administration, envisions a broader arrangement in which the U.S. government would hold 5% stakes in major AI developers through a sovereign wealth fund. This would include companies such as Anthropic, Google, and Meta, though it remains unclear whether these firms would agree to similar terms. The Trump administration has not yet engaged with Anthropic on the matter, according to a source familiar with the situation. Altman first introduced the idea to the Trump administration in early 2025, and the proposal has been part of ongoing negotiations for over a year. The push for government stakes comes amid heightened concerns over cybersecurity risks associated with AI models and the growing competitiveness of Chinese open-source alternatives. These models, which are often cheaper and equally capable, have intensified pressure on U.S. firms to align with regulatory demands. Anthropic, for instance, recently disabled access to its most advanced Mythos and Fable models to comply with export control directives.#google #trump_administration #anthropic #openai #sam_altman

Meta Compute: Everyone Wants To Be A Neocloud Meta’s aggressive expansion into the neocloud market has sparked intense speculation, with analysts and investors debating the company’s strategic moves. Despite initial market reactions that suggested a slowdown in datacenter investments, the reality is that Meta’s procurement of compute resources is accelerating. The company has already contracted over 5GW of capacity across cloud and colocation services in the first half of 2024, with self-built datacenters adding to this figure. Analysts at SemiAnalysis argue that the notion of “half of US datacenters being delayed” is misleading, as just two of Meta’s datacenters represent more than half of the current construction pipeline. The core of Meta’s strategy lies in its ability to monetize vast compute resources through multiple high-value use cases. One key area is scaling its recommendation systems (RecSys) by more than 10x in complexity to drive revenue growth. This requires both inference and training compute, enabling more sophisticated generative advertising. Additionally, Meta is reportedly in final talks with Anthropic to access private instances of Claude, akin to Amazon’s Bedrock or Google’s Vertex. This partnership could power internal tools and external SaaS platforms leveraging Frontier AI agents, with plans to launch a token-as-a-service endpoint. Meta’s ambitions extend beyond internal use. The company is positioning itself to replicate Elon Musk’s “SpaceX-type” deals, where large-scale on-demand compute is sold at premium pricing. A potential $10B deal with Anthropic could kickstart a flywheel effect, leveraging Meta’s network to scale compute offerings.#meta #anthropic #coreweave #semi_analysis #neocloud

Anthropic Launches Claude Sonnet 5 as a Cheaper Way to Run Agents Anthropic has introduced Claude Sonnet 5, a new midsize model designed to offer agentic capabilities at a lower cost compared to its predecessors. The release marks a shift in the competitive landscape of foundation models, where agentic functionality—allowing systems to plan, use tools, and operate autonomously—is now a baseline expectation. Sonnet 5 is positioned as a more affordable alternative to Anthropic’s larger Opus 4.8 model, with pricing adjustments set to take effect starting June 30, 2026. The model’s agentic features are highlighted as a key differentiator. Anthropic claims Sonnet 5 can perform tasks that previously required more expensive, larger models, such as autonomous planning and tool usage. This aligns with trends observed in other major AI companies, including OpenAI and Google, which have also emphasized agentic capabilities in their recent releases. OpenAI’s GPT-5.6 Sol and Google’s Gemini 3.5 Flash are similarly framed as tools that reduce human oversight in complex workflows. Sonnet 5’s pricing structure is a significant factor in its appeal. Initially priced at $2 per million input tokens and $10 per million output tokens through August 31, 2026, the model becomes more expensive afterward, with input token costs rising to $3 per million. This pricing makes Sonnet 5 cheaper than Opus 4.8, GPT-5.5, and Gemini 3.1 Pro, though it remains more costly than Gemini 3.5 Flash. Anthropic emphasizes that the model’s affordability, combined with its performance, provides developers with a flexible option for balancing cost and efficiency. Performance benchmarks indicate that Sonnet 5 outperforms its predecessor, Sonnet 4.6, in several areas. On agentic coding tasks, Sonnet 5 scores 63.2%, compared to Opus 4.8’s 69.#anthropic #opus_4_8 #claude_sonnet_5 #gpt_5_6_sol #gemini_3_5_flash

Introducing Claude Sonnet 5 Anthropic has launched Claude Sonnet 5, the latest iteration of its agentic AI model series, designed to deliver advanced capabilities in planning, tool integration, and autonomous execution at a lower cost than previous models. This release marks a significant step forward in the agentic AI era, bridging the gap between Sonnet-class and Opus-class models while offering competitive pricing and improved performance. Sonnet 5 is positioned as a substantial upgrade over its predecessor, Sonnet 4.6, with notable improvements in reasoning, tool use, coding, and knowledge work. Performance evaluations show it closely rivals Opus 4.8, a more broadly capable model, but at a significantly lower cost. The model’s agentic capabilities are highlighted through comparisons with earlier versions, demonstrating its ability to handle complex tasks such as multi-step software engineering workflows, cybersecurity assessments, and automated workflows with greater efficiency. Pricing for Sonnet 5 is structured to make it accessible across all user tiers. It is the default model for Free and Pro plans and is available to Max, Team, and Enterprise users. During its introductory launch period through August 31, 2026, Sonnet 5 is priced at $2 per million input tokens and $10 per million output tokens, after which the rates will increase to $3 per million input tokens and $15 per million output tokens. Developers can access the model via the Claude API using the identifier claude-sonnet-5. The model’s performance is benchmarked against Sonnet 4.6 and Opus 4.8 in evaluations such as agentic search (BrowseComp) and computer use (OSWorld-Verified). Sonnet 5 is described as a strict improvement over Sonnet 4.6, offering a balance of cost and performance that makes it a compelling alternative to Opus 4.#anthropic #mythos_preview #claude_sonnet_5 #sonnet_4_6 #opus_4_8