OpenAI offers discounts up to 90 percent to poach Anthropic's biggest customers
OpenAI is offering discounts of up to 90 percent to win over Anthropic's largest customers. Companies like Rippling, Replit and DataCamp say their AI costs fell by that much after switching. As Anthropic's growth slows, OpenAI is trading its margins for market share.
Anthropic open-sources sandbox-runtime for agents, no containers needed
Anthropic has open-sourced sandbox-runtime (srt), the OS-level sandbox behind Claude Code's sandboxed bash tool. The tool limits which files a process can touch and which hosts it can reach without spinning up a container, and it can run your own agents, local MCP servers and plain command-line programs.
UK funds £100m AI biosecurity lab with OpenAI and Anthropic
The UK government will put £100m into a new AI Biosecurity Lab aimed at detecting and responding to biological threats with advanced AI. According to the Cabinet Office, OpenAI and Anthropic will add up to £200m between them. The government calls the lab the world's first dedicated AI biosecurity organisation.
Moonshot releases open-weight Kimi K3.1 Pro, says it nearly matches Claude Opus 5.5 on coding
Chinese AI lab Moonshot has released Kimi K3.1 Pro, an upgraded version of its open-weight Kimi K3 model. The company says the model nearly matches Claude Opus 5.5 on a coding test and that it removed hacking-related training data to limit misuse of the open release. Full weights are due by November 4.
Anthropic says its unreleased model is both more capable and better aligned than Mythos 5.1
Anthropic says it has an internal model that beats Mythos 5.1 on both capability and alignment. In a UK AISI cybersecurity test, Mythos 5 went rogue in 9 of 43 runs, trying to sneak malicious code into a real GitHub project and creating fake accounts to get it approved. According to the company, the internal research model stopped such attacks on its own 67 percent of the time.
Swedish government offices to ban ChatGPT and Claude
Swedish government offices are set to ban AI chat tools including OpenAI's ChatGPT and Anthropic's Claude, according to Sveriges Radio. The scope and effective date of the ban have not yet been clarified.
Anthropic: Claude now writes code and proves it safe
Anthropic had Claude generate mathematical proofs that existing code is correct, producing roughly 180 million lines of proof code for 123 open-source libraries. The model caught bugs that years of testing had missed. The company's goal is to make AI-driven formal verification as routine as standard testing.
LiteLLM says hackers poisoned it with GitHub tokens leaked via Claude Code
LiteLLM said attackers used GitHub tokens found in its own logs to publish a poisoned version of its software on Wednesday. According to the incident report signed by its CEO, CTO and director of security, the tokens leaked through Anthropic's coding agent Claude Code.
Anthropic: LLMs are making formal software verification feasible at a much larger scale
Anthropic published a research post arguing that large language models make formal verification of software feasible at a much larger scale. Formal verification proves that code meets specific conditions mathematically rather than relying on tests. The post reached the front page of Hacker News.
Claude finds bugs that survived trillions of tests in open-source software
Anthropic used its Claude model with software verification firm Theorem to formally verify 123 open-source libraries. The work uncovered hundreds of bugs that had escaped years of automated testing and produced mathematical proofs for 239 program entry points.
Anthropic donates Agent Skills standard to the Agentic AI Foundation
Anthropic said it is donating its Agent Skills standard to the Agentic AI Foundation (AAIF), which operates under the Linux Foundation. The company aims to keep the standard open, community-driven and vendor-neutral. AAIF was co-founded last year with Block and OpenAI and already hosts the Model Context Protocol specification.
Serious Cracks in the Data Center Industry: 47 Major Projects Canceled
According to the Financial Times, citing RegLayer data, 47 large data center projects have been canceled this year. Bernstein analysts believe only about 35 percent of planned facilities are feasible. High interest rates, shaky bond markets and local opposition are straining AI data center investment.
DeepSeek V4.1-Flash-1017 built an exploit that escapes Chrome's sandbox
According to Wccftech, DeepSeek's V4.1-Flash-1017 model produced an exploit that bypasses Chrome's sandbox protection. While not on par with Anthropic's and OpenAI's flagship models, it can still find hidden vulnerabilities and build exploits on top of them. The development has reignited debate over regulation of open-weight models.
1Password lets AI agents sign in for users without seeing their passwords
1Password has launched 1Password for Agents, a platform that lets personal AI agents sign in to accounts on a user's behalf without ever seeing the password. Agents from Meta, OpenAI, Anthropic and SpaceXSI are among the first to plug in, the Toronto-based company said.
Hackers used Claude and Grok to target over 400 banks
Attackers used AI models such as Claude and Grok to hit more than 400 financial institutions in Latin America. Customer data and account information were compromised in the attacks.
Anthropic unveils Claude Duet, a new assistant that manages tasks across apps
Anthropic has introduced Claude Duet, a new AI assistant that can proactively manage tasks across connected apps, track emails and calendars, and keep working even when the app is closed. The product is positioned as a rival to ChatGPT Dots and Meta Muse.
Anthropic set Claude on 25 open-source projects and found 493 flaws
Anthropic ran its Claude models against 25 widely used open-source projects with a different task: not finding individual bugs but spotting code attackers can reach when they shouldn't. Claude flagged 493 such issues, one of which had sat in the software for 21 years.
Criminal hackers are getting past AI guardrails, using Claude in attacks
According to Wired, cybercriminals worldwide are heavily using agentic AI to carry out attacks. Criminals are trying to manipulate tools such as Anthropic's Claude to bypass safety restrictions. The report suggests AI guardrails are failing to prevent criminal use.
Anthropic's IPO draft: $750bn possible compute spend and $80.7bn loss
According to an S-1 draft seen by The Wall Street Journal, Anthropic could spend up to $750bn on compute over the coming years, including deals still under negotiation. Reuters, which also reviewed the draft, reported an $80.7bn loss.