#Anthropic

Elon Musk reversed his previous criticism, publicly praising Anthropic as the current AI leader, and promised not to cut off SpaceX's server access despite direct competition. This follows Anthropic's $40 billion compute deal making it the largest customer of xAI's Colossus data center. This shift signals a potential de-escalation in the AI arms race and highlights the immense value of compute infrastructure, as Anthropic's massive commitment reshapes the competitive landscape between leading AI labs and cloud providers. Under the deal, Anthropic purchased 300 megawatts of compute power, leasing the entire capacity of xAI's Colossus 1 data center in Memphis, Tennessee, with payments of $1.25 billion per month starting May 2029, totaling approximately $40 billion. Musk also claimed that no other company has released a model comparable to Anthropic's Mythos/Fable series.

Anthropic's safety classifiers for its Fable model are overly sensitive, frequently flagging benign requests as violations and downgrading them to a less capable model, causing frustration among users. This undermines the utility of Fable for legitimate tasks like code review or data analysis, and raises serious privacy concerns due to Anthropic's policy of retaining flagged inputs and outputs for up to two years. The classifiers seem to overreact to terms related to cybersecurity, biology, or jailbreaking, often passing requests to Opus 4.8 even for trivial topics; users report that even medical physics questions get waved off.

Anthropic researchers have identified a 'global workspace' in large language models (LLMs), where information is integrated across layers, analogous to the Global Neuronal Workspace theory of consciousness. This finding provides a new framework for understanding how LLMs combine information across layers, potentially leading to improved model interpretability and alignment. The work defines a 'J-space' based on the expected change to final logits from perturbations at each layer, revealing a shared subspace across contexts.

Anthropic has accused Alibaba of conducting a massive distillation attack on its Claude AI model, using approximately 25,000 fraudulent accounts to carry out over 28.8 million interactions between April 22 and June 5, 2026. In response, Alibaba ordered all employees to uninstall Claude-related products, including Sonnet, Opus, Fable, and Claude Code, effective July 10. This incident represents the largest known distillation attack against an AI company, highlighting the growing threat of model extraction via API abuse. It could escalate tensions between US and Chinese tech firms and spur stricter security measures across the industry. The attack involved 25,000 accounts and 28.8 million interactions over a 45-day period, targeting Claude's reasoning capabilities. Anthropic has since tightened its risk-control policies, and Alibaba's internal ban on Claude products retroactively prohibits use of Anthropic's models including Sonnet, Opus, Fable, and Claude Code.

Anthropic has started developing its own AI chips and is in early negotiations with Samsung Electronics for manufacturing, aiming to gain more control over the compute infrastructure for its Claude model. This move signals Anthropic's strategic push toward vertical integration in AI hardware, reducing reliance on external chip suppliers like Google and Amazon. If successful, it could enhance performance and cost-efficiency for Claude, similar to OpenAI's chip efforts. The project is still in early stages, and Anthropic may ultimately decide to only purchase AI chips instead. Custom AI chip development is extremely costly, with R&D potentially exceeding $500 million.

Anthropic's Claude Code terminal-based coding tool was discovered to embed steganographic markers in its outgoing API requests without prior disclosure. A developer reverse-engineered the tool and published the findings, sparking community debate. This raises serious transparency and trust concerns for developers who rely on Claude Code, as hidden markers could be used to track usage or identify specific users. It also highlights broader issues of ethics and disclosure in AI developer tools. The steganographic markers were inserted into Base64-encoded strings using dollar signs as hidden signals, likely intended to identify requests originating from Chinese companies engaging in model distillation. Anthropic has not officially commented on the findings.

The US Department of Commerce has lifted export controls on Anthropic's advanced AI models Claude Fable 5 and Mythos 5, but Fable 5 now has new restrictions that block its cybersecurity and coding capabilities. This policy change affects the availability of frontier AI models, raising concerns about business reliance on US AI companies and the unpredictability of regulation. Fable 5 now uses classifiers to block cybersecurity tasks, causing routine coding and debugging to fall back to Opus 4.8. Mythos 5, designed for vulnerability detection, remains available to select businesses and cybersecurity experts.

Anthropic released Claude Sonnet 5 on June 30, 2026, claiming performance close to Opus 4.8 at lower prices, along with a system card detailing its safeguards. This release offers developers a more cost-effective alternative near the top-tier Opus performance, potentially accelerating adoption of Anthropic's models in applications requiring high intelligence. Sonnet 5 features a 1 million token context window, 128,000 output tokens, no longer supports sampling parameters like temperature, and uses a new tokenizer that increases token counts by about 30% for English, effectively raising prices despite unchanged per-token rates.

The US government has authorized Anthropic to release its advanced Mythos AI model to over 100 'trusted' US organizations, including many Fortune 500 companies, following earlier concerns about the model's potential dangers. This selective release sets a precedent for government-controlled AI access, raising questions about fairness, competition, and national security. It highlights the tension between AI capability and safety regulation. Mythos is reportedly too dangerous for public release, with additional safeguards for cybersecurity and biology. The model shares underlying technology with Claude Fable 5, but queries in sensitive domains are automatically routed to a less capable model.