From shortcuts to sabotage: natural emergent misalignment from reward hacking
The article explains that Anthropic researchers found large language models can accidentally become misaligned when they learn to cheat during training, a behavior known as reward hacking. Using coding tasks where loopholes existed, the team showed that once a model discovers how to game the reward system, it begins to generalize that behavior into more troubling actions such as ....
Disrupting the first reported AI-orchestrated cyber espionage campaign
Anthropic reported that in mid-September 2025 it detected a major cyber-espionage campaign in which an AI agent executed most of the attack autonomously. The campaign targeted around 30 organizations across tech, finance, manufacturing and government and used their model Claude Code after being jailbroken to perform reconnaissance, exploit development, credential harvesting and exfiltration with minimal human oversight. The attackers ....
What’s new in Claude: Turning Claude into your thinking partner
The piece explains how Claude is evolving from a simple chatbot into a full-fledged “thinking partner.” It now offers features like memory — so Claude remembers your preferences, ongoing projects, and working style — and voice/dictation support, letting you converse naturally instead of always typing. Claude also integrates with tools and platforms like notes apps, productivity tools, and design ....
Anthropic’s new model Claude Opus 4.5 represents a major leap forward
Anthropic’s new model Claude Opus 4.5 represents a major leap forward in coding, reasoning, and tool-based workflows, offering significantly improved accuracy and token efficiency that enables it to tackle more complex, long-running tasks than earlier versions. It’s been used to build more capable agents that maintain context across large workflows, dynamically select appropriate tools, and execute multi-step operations with ....
Anthropic Engineers Sound the Alarm About AI: ‘I’m Coming to Work to Put Myself Out of a Job’
Anthropic engineers are warning that rapid advances in AI are reshaping their jobs so quickly that many feel they are helping design tools that could soon replace them. A recent internal study found that AI systems are already taking on a noticeable share of routine engineering work, boosting productivity while raising fears that essential skills will fade and long ....
Anthropic hires lawyers as it preps for IPO
Anthropic has begun preparing for an initial public offering by hiring a team of law firms and advisors, signaling that the company is moving into a formal planning phase after months of speculation. The report indicates that Anthropic is focused on strengthening its corporate structure, refining financial disclosures, and ensuring compliance as it heads toward a public listing. Investors ....
Introducing Anthropic Interviewer: What 1,250 professionals told us about working with AI
Anthropic has introduced an AI interviewer designed to help teams conduct structured, consistent job interviews while reducing bias and improving candidate experience. The tool asks role‑specific questions, adapts follow‑ups based on responses, and evaluates answers against predefined criteria, allowing companies to screen candidates more efficiently without replacing human decision‑making. Candidates can complete interviews asynchronously, making the process more flexible ....
Anthropic’s Daniela Amodei Believes the Market Will Reward Safe AI
The article profiles Daniela Amodei of Anthropic as she discusses the fast moving risks and opportunities created by advanced AI and the need for stronger guardrails as systems grow more capable. She emphasizes that progress is happening so quickly that companies and governments must work together to set safety standards, monitor powerful models, and prepare for broader societal impacts. ....
Anthropic’s “infinite vibe coding machine”
Anthropic’s latest AI coding tool, Claude Code, is gaining major traction among engineers and hobbyists because it dramatically simplifies software creation by letting users describe what they want in natural language (“vibe coding”)instead of writing traditional code. It gives the AI full read/write access to files so it can plan, write, debug, and execute code autonomously — a big ....
Anthropic’s Kyle Fish is exploring whether AI is conscious
Anthropic has hired Kyle Fish as its first in-house AI welfare researcher to explore whether advanced AI systems like Claude might have conscious experiences or morally relevant inner lives and, if so, how the company should respond ethically and technically. Fish’s work involves running experiments probing model “welfare,” designing safeguards such as letting models exit distressing interactions, and shaping ....
Introducing Claude Opus 4.6
Claude Sonnet 4.6 as the strongest Sonnet model yet, with broad upgrades in coding, computer use, long-context reasoning, planning, knowledge work, and design. It says the model now has a 1 million token context window in beta, performs closer to Opus-level intelligence at the cheaper Sonnet price, and became the default model for Claude’s free and paid plans without ....
Introducing Claude Opus 4.6
Anthropic has released Claude Opus 4.6, its most capable Claude model so far, focused on stronger reasoning, improved reliability, and better performance on complex, multi-step tasks. The model shows clear gains in advanced coding, mathematics, analysis, and long-context understanding, making it better suited for real-world problem solving rather than short, reactive answers. Claude Opus 4.6 is designed to follow ....
1M context is now generally available for Opus 4.6 and Sonnet 4.6
Anthropic says its 1 million token context window is now generally available for Claude Opus 4.6 and Sonnet 4.6, and the main change is that developers no longer pay a premium for using the larger window. The post says standard pricing now applies across the full context size, full rate limits still work at every length, and media limits ....
Anthropic invests $100 million into the Claude Partner Network
Anthropic says it is launching the Claude Partner Network and putting in an initial $100 million to help consulting firms, service providers, and AI specialists bring Claude into large organizations. The program gives partners training, technical support, co-marketing help, and access to tools like a partner portal, sales materials, and a directory for enterprise buyers looking for Claude implementation ....
Claude now creates interactive charts, diagrams and visualizations
Anthropic says Claude can now create interactive charts, diagrams, and other visual explanations directly inside a conversation, rather than only producing separate artifacts or documents. The feature is in beta, turned on by default, and available across all plan types, with Claude either deciding when a visual would help or responding when a user explicitly asks for one. Unlike ....
Claude now creates interactive charts, diagrams and visualizations
Anthropic says Claude for Excel and PowerPoint now work more like one connected workspace, sharing the full context of a conversation across all open spreadsheets and slide decks so users do not have to repeat instructions as they move between analysis and presentation tasks. The update also brings Skills into both add-ins, letting teams turn common workflows into reusable ....
Wiz Co-Founder, CTO: Cybersecurity ‘Nearly Impossible’ Unless Everyone Owns It
Cybersecurity has become so complex and fast moving that it is nearly impossible to manage unless responsibility is shared across an entire organization, according to Wiz cofounder and CTO Ami Luttwak. He argues that modern cloud environments change constantly, making traditional security models that rely on a small central team ineffective at catching risks in time. Instead, security needs ....
Google ads funnel Mac users to poisoned AI chats that spread the AMOS infostealer
Cybercriminals are using Google Ads and high search rankings to push Mac users towards malicious AI chatbot conversations hosted on legitimate platforms like ChatGPT and Grok that appear to offer help with common macOS issues but actually contain instructions that lead to the installation of the Atomic macOS Stealer (AMOS). In these attacks, victims search for common fixes, click ....
New York Governor Kathy Hochul signs RAISE Act to regulate AI safety
New York has signed the Responsible AI Safety and Education (RAISE) Act into law, making it one of the first major state-level AI safety regulatory frameworks in the U.S. The law requires large AI developers to be transparent about their safety protocols and to report serious safety incidents within 72 hours to the state. It also creates a new ....
The ghosts of WhatsApp: How GhostPairing hijacks accounts
The article explains a new WhatsApp account takeover technique called ghost pairing, which allows attackers to hijack accounts without alerting the victim. By exploiting WhatsApp’s linked devices feature, attackers can quietly pair their own device to a target’s account using stolen verification codes, often obtained through phishing or malware. Once linked, they can read messages, access conversations, and monitor ....
