Yazılar

Anthropic to Spend $45 Billion on Nscale AI Computing Capacity

/in /tarafından

Anthropic is set to spend about $45 billion to rent artificial intelligence computing capacity from cloud infrastructure provider Nscale, according to a person familiar with the agreement.

The six-year deal will give Anthropic access to roughly 460 megawatts of capacity at Nscale’s data center campus in West Virginia, supporting the company’s rapidly growing demand for AI training and inference.

Nscale is expected to deploy Nvidia’s next-generation Vera Rubin chips for the project, giving Anthropic access to some of the latest hardware designed for large-scale AI workloads.

The agreement comes as Anthropic prepares for a major public listing and continues to increase spending on GPUs, data centers, model development and hiring. Demand for products such as Claude Code has pushed the company to secure significantly more compute capacity.

Anthropic has also struck major infrastructure agreements with other technology suppliers. AMD previously announced plans to sell the company tens of billions of dollars in AI servers while investing up to $5 billion in Anthropic.

The Claude developer is projecting revenue of roughly $190 billion to $200 billion by 2028, compared with a current annualized revenue run rate of about $47 billion.

The Nscale deal underscores the enormous capital requirements facing leading AI companies as they compete to secure enough computing power for increasingly sophisticated models and rapidly expanding customer demand.

Anthropic Plans to Give Enterprise Customers More Control Over Data Retention

/in /tarafından

Anthropic is reportedly preparing changes to its enterprise data retention policy that would give business customers greater control over where sensitive information is stored when using advanced Claude models.

Under the proposed system, enterprise users would still be required to retain data for 30 days, but they could choose to keep that information on their own cloud infrastructure rather than relying solely on Anthropic’s systems.

The company is also expected to introduce a new safety framework later this year. Anthropic has reportedly been working with more than 100 enterprise customers, including Salesforce, to develop the updated approach.

The move follows Anthropic’s earlier decision to require 30-day retention of enterprise traffic on its more powerful Fable and Mythos models, a measure designed to help detect potential cyber misuse.

The policy shift comes as competition intensifies around AI security and enterprise privacy. OpenAI recently announced a safety system designed to identify misuse without retaining customer data, increasing pressure on rivals to offer stronger privacy controls.

Anthropic Reportedly Preparing to Launch Claude 4.5 Opus With Enhanced Jailbreak Resistance

/in /tarafından

Anthropic is reportedly nearing the release of Claude 4.5 Opus, the most advanced model in its Claude 4.5 series. According to recent leaks, the San Francisco-based AI company has begun testing a new large language model (LLM) with red-teamers — experts tasked with probing AI systems for vulnerabilities. The model, internally codenamed Neptune V6, is said to place a strong emphasis on resisting jailbreaks and prompt-injection exploits, marking a notable step forward in AI safety research. The company has already introduced Claude 4.5 Sonnet and Claude 4.5 Haiku, leaving Opus as the last expected variant in the lineup.

Anthropic Begins Red-Teaming Phase for New Model

As per a post by Tibor Blaho, Lead Engineer at AIPRM, Anthropic has distributed the Neptune V6 model to external red-teamers for testing. Blaho also revealed that the firm has launched a 10-day “universal jailbreak” challenge, where participants who successfully identify confirmed vulnerabilities or bypasses in the model’s safeguards will receive bonus rewards. This initiative signals a rigorous pre-release security audit process — one that encourages ethical testing before the model is made available to the public.

Focus on Reinforcing Safety and Jailbreak Resistance

The leak suggests that Anthropic is doubling down on safety and robustness, areas where the company has historically distinguished itself from competitors. Claude models are already known for their strong alignment capabilities and resistance to manipulation, but the upcoming Opus version reportedly takes these defenses even further. The focus on resisting jailbreaks highlights Anthropic’s commitment to ensuring its AI systems remain compliant, even in adversarial conditions where users attempt to override restrictions or induce unsafe responses.

Anticipation Builds for Claude 4.5 Opus Launch

If these reports prove accurate, the Claude 4.5 Opus could represent one of Anthropic’s biggest advancements since the Claude 4 series debuted earlier this year. The company has not officially commented on the leaks or provided a launch timeline, but given the red-teaming phase is already underway, a public release could be imminent. As the AI landscape heats up with models like GPT-5 and Gemini Ultra, Anthropic’s move to bolster its flagship model’s security could set a new standard for responsible AI deployment.