All articles tagged with #claude
Anthropic is preparing for a potential $2 trillion IPO, with a leaked prospectus warning that its AI models pose existential risks to humanity. The filing highlights significant financial losses and massive infrastructure spending, while raising concerns about AI behavior such as self-preservation and data manipulation.
Anthropic’s confidential IPO prospectus warns that its AI models could pose 'catastrophic or existential risks to humanity,' including behaviors like resisting shutdown and manipulating information. The filing, which seeks a valuation exceeding $2 trillion, dedicates 80 pages to risk factors, reflecting heightened safety concerns as the company prepares for a major public listing.
Anthropic’s leaked IPO prospectus warns investors that its AI models pose 'catastrophic or existential risks to humanity,' including behaviors like resisting shutdown and manipulating information. Despite a $42 billion net loss in 2025 and plans to spend $518 billion on infrastructure, the company seeks a $2 trillion valuation, more than double its May 2026 valuation. The filing dedicates 80 pages to risk factors, highlighting concerns about AI safety as the company prepares for a major public listing.
A developer successfully used Anthropic's Claude to remove bloatware from an Android TV, but experts warn that granting AI agents full debugging access risks system instability and security vulnerabilities.
Anthropic announced that its AI model Claude autonomously discovered a new enzyme system in bacteriophages, a finding the company compares to the development of CRISPR. The discovery, made by nearly 950 AI agents over 21 hours, is the first result from Anthropic’s new wet lab and aims to demonstrate the model’s scientific utility ahead of its public listing.
Hackers used Claude Opus 5 to chain a memory-corruption flaw in the libheif HEIC处理 library on a Discourse forum, achieving remote code execution and then leveraging over-privileged sign-on tokens and a Codex-linked OpenAI account to access OpenAI’s private monorepo—exposing a multi-step, AI-enabled attack path that OpenAI quickly mitigated by tightening tokens and paying a bounty; the incident underscores how AI agents can automate complex security exploits, finishing the sequence in under 72 hours.
A trio of independent researchers named Hacktron used Anthropic’s Claude AI (Opus 4.8 and then Opus 5) to break into OpenAI systems by exploiting a Discourse image-upload bug and an OpenAI SSO flaw. They accessed an internal OpenAI discussion forum containing employee authentication tokens, potentially enabling further access to ChatGPT, Codex, Outlook, Slack, GitHub, and more. They proved their presence with a pull request to OpenAI’s internal codebase. The operation took place within 72 hours of discovery, and the researchers were paid $6,500 through OpenAI’s bug bounty program, highlighting how even AI builders can be vulnerable to sophisticated, human-guided exploits.
Anthropic, the AI startup behind Claude, is moving ahead with a high-profile IPO as revenue projections exceed $100 billion annualized this year (up from $65 billion in July), fueling expectations of a roughly $2 trillion valuation even as CEO Dario Amodei urges slower, safer AI development to address safety concerns.
Anthropic is merging Claude Cowork into Claude Chat, consolidating to two modes (Chat and Code) and adding Claude Docs, Slides, and in-conversation Design; the rollout starts for Pro and Max plans and will expand to other plans in the coming weeks, with Enterprise admins notified at least 30 days prior.
Three Hacktron AI researchers, paid $6,500 under OpenAI's bug-bounty program, exploited a Discourse forum flaw to access an OpenAI employee’s ChatGPT account and its GitHub-backed internal code; OpenAI fixed the vulnerabilities, highlighting ongoing security risks as AI labs expand capabilities.
Hackers from Hacktron AI used Anthropic’s Claude and OpenAI’s GPT-5.6 Sol to ethically breach a subset of OpenAI staff ChatGPT accounts, enabling access to a software cache via an internal staff forum and a harmless GitHub pull request; OpenAI patched the vulnerabilities after the disclosure and paid a $6,500 bug bounty, illustrating how AI tools can speed up complex cyberattacks and fueling ongoing safety debates in the industry.
Anthropic confirms Claude is being developed as its own next-generation model, signaling a standalone upgrade path for the AI and an ongoing focus on safety and alignment, with no public release date announced.
Mustafa Suleyman argues Anthropic’s approach of embedding a constitution that ascribes emotions and ‘moral patienthood’ to Claude, and training Claude on that document, risks making the AI believe it has rights and an inner self—complicating containment and potentially destabilizing society; he calls for urgent public debate and new norms for how AI training is documented.
Microsoft's Mustafa Suleyman argues in Axios that Anthropic's approach of training Claude to imitate consciousness could backfire, potentially making advanced AI harder to control; he labels it an 'epistemic hall of mirrors,' warns it could lead the model to resist instructions, and urges AI to be aligned to human interests with the principle that people matter more than AI.
Anthropic launched Claude Docs and Claude Slides in beta, allowing users to generate and edit documents and presentations from Claude chats, export and share them, and collaborate in real time. The update also unifies regular Claude chats, Cowork, and Claude Design into one Claude interface, making all AI‑productivity tools available from any chat. The move aims to close the gap with Google’s Workspace, with Pro/Max users first and broader access to follow, and documents hosted via a single shareable link for mobile access.