Anthropic has released its latest AI model, Claude Sonnet 4.5, which it claims is the world's best for coding and autonomous tasks. This new model can operate independently for up to 30 hours and shows significant improvements in computer use, cybersecurity, and financial services. Beta testers like Canva have noted its ability to handle complex, long-context tasks. Anthropic is also empowering developers with tools like the Claude Agent SDK to build their own AI agents. Claude Sonnet 4.5 is now available via API at the same price as its predecessor and is also being integrated into platforms like GitHub Copilot for users with specific subscription plans. Meanwhile, in the realm of AI misuse, former President Donald Trump shared and later deleted an AI-generated video promoting the 'MedBed' conspiracy theory, drawing criticism from California Governor Gavin Newsom. Cyberattackers are also leveraging AI for more sophisticated spam, malicious code, and phishing lures, while also targeting AI systems themselves with malicious prompts and corrupting data used by AI security tools. Experts like UCLA's chief data and artificial intelligence officer highlight AI's role in enhancing phishing and social engineering attacks, emphasizing the need for AI-driven defense strategies. On a different note, Epidemic Sound has introduced an AI tool called Adapt, which modifies existing music tracks rather than generating new ones, aiming to support creators and increase royalty payouts for artists. Amazon is investing an additional $1.9 billion into its Delivery Service Partner program, focusing on increasing driver pay and using AI-powered tools for enhanced safety and efficiency. DeepSeek has launched its experimental model, DeepSeek-V3.2-Exp, which is more efficient and better at processing long text sequences, accompanied by a 50% API price cut. Tanka has launched EverMemOS, an AI and memory-native operating system designed to provide persistent, searchable memory for AI applications. Google's CTO of technology and society, AgĆ¼era y Arcas, posits that human intelligence is fundamentally computational, akin to AI, focusing on prediction and influence as key aspects of intelligence.
Key Takeaways
- Anthropic's new AI model, Claude Sonnet 4.5, excels in coding and autonomous tasks, capable of operating independently for up to 30 hours.
- Claude Sonnet 4.5 is being integrated into GitHub Copilot for select subscribers, enhancing tool orchestration and specialized tasks.
- Former President Donald Trump shared and deleted an AI-generated video promoting the 'MedBed' conspiracy theory, drawing criticism for spreading misinformation.
- Cyberattackers are using AI to create more sophisticated malicious content and are exploiting AI systems through poisoned prompts and data corruption.
- AI is enhancing phishing and social engineering attacks, necessitating AI-driven defense strategies in cybersecurity, as noted by UCLA's AI expert.
- Epidemic Sound's AI tool, Adapt, modifies existing music tracks to aid creators and increase artist royalties, rather than generating new songs.
- Amazon is investing $1.9 billion in its delivery partners to boost driver pay and implement AI for improved safety and efficiency.
- DeepSeek has released its experimental AI model DeepSeek-V3.2-Exp, featuring improved efficiency for long text processing and a 50% API price reduction.
- Tanka has launched EverMemOS, an AI and memory-native operating system designed to offer unlimited, searchable memory for AI applications.
- Google's CTO suggests human intelligence is computational, similar to AI, defining intelligence by the ability to predict and influence the future.
Anthropic's Claude Sonnet 4.5 boosts AI coding and autonomous tasks
Anthropic has released its new AI model, Claude Sonnet 4.5, which it claims is the best in the world for coding and AI agents. This model can operate autonomously for up to 30 hours, a significant improvement from previous versions. Claude Sonnet 4.5 also shows enhanced abilities in using computers, cybersecurity, and financial services. Beta testers like Canva have reported that the model helps with complex, long-context tasks. Anthropic is also providing developers with tools to build their own AI agents using this new model.
Anthropic's Claude Sonnet 4.5 leads AI coding and autonomous work
Anthropic has launched Claude Sonnet 4.5, its latest AI model, which the company calls the world's best for coding. The new model can work independently for up to 30 hours and shows significant improvements in computer use. Anthropic states that Sonnet 4.5 delivers top results on software engineering benchmarks and is designed to maintain the company's lead in AI coding. The Claude Agent SDK is also being released to provide developers with tools for creating customized AI agents.
Anthropic's Claude Sonnet 4.5 is a smarter AI colleague
Anthropic has released its newest AI model, Claude Sonnet 4.5, which is designed to be a more capable assistant for various tasks. The company states this model excels in coding, computer use, and meeting practical business needs, particularly in fields like cybersecurity, finance, and research. Claude Sonnet 4.5 can operate autonomously for up to 30 hours, a substantial increase from earlier models. Anthropic aims for this model to be the default choice for users due to its improved intelligence and safety features.
Claude Sonnet 4.5 now in GitHub Copilot preview
Anthropic's advanced AI model, Claude Sonnet 4.5, is now available in a public preview for GitHub Copilot users with specific subscription plans. This model offers major upgrades for tool orchestration, context editing, and specialized tasks. Users can select Claude Sonnet 4.5 within Copilot Chat across various platforms like Visual Studio Code and GitHub.com. The update also includes access for Copilot CLI users, with gradual rollout expected for all users.
Anthropic's Claude Sonnet 4.5 works autonomously for 30 hours
Anthropic has launched Claude Sonnet 4.5, its latest AI model, which can work autonomously for up to 30 hours on complex tasks. The company reports significant improvements in coding, finance, and cybersecurity capabilities compared to previous versions. Sonnet 4.5 is priced the same as its predecessor, Claude Sonnet 4.0. Anthropic is also providing developers with access to Claude Code's building blocks to simplify the creation of AI agents.
Anthropic introduces Claude Sonnet 4.5 AI model
Anthropic has released Claude Sonnet 4.5, its most advanced AI model yet, excelling in coding, building agents, and computer use. The model can maintain focus on complex tasks for over 30 hours. New features in Claude apps include checkpoints for saving progress and a refreshed terminal interface. Developers can use the Claude Agent SDK to build their own AI agents. Claude Sonnet 4.5 is available now via API at the same price as Claude Sonnet 4.
Trump shares deleted AI video about 'miracle hospital beds'
California Governor Gavin Newsom criticized former President Donald Trump for sharing an AI-generated video promoting a conspiracy theory about 'miracle hospital beds.' The video, posted on Trump's platform and later deleted, falsely claimed these beds would soon be available to all Americans. Newsom called the post 'AI-generated slop' and stated that Trump 'has lost it.' The 'MedBed' conspiracy theory suggests secret advanced healing technology is being withheld from the public.
Trump shares AI video promoting 'MedBed' conspiracy
Donald Trump shared an AI-generated video on Truth Social that promoted the conspiracy theory surrounding 'MedBeds,' futuristic medical pods claimed to cure diseases. The video, which was later deleted, suggested Americans would receive 'MedBed cards' for access to advanced healthcare. The concept of MedBeds is popular in far-right and QAnon circles, alleging that secret healing technology is being hidden from the public. California Governor Gavin Newsom mocked Trump for sharing the video.
Trump shares AI deepfake video promoting fictional 'MedBed'
Former President Donald Trump shared an AI-generated video on Truth Social featuring a deepfake of himself promoting a nonexistent product called a 'MedBed.' The video, presented as a Fox News segment, claimed universal access to this supposed healing technology. Conspiracy theories surrounding MedBeds suggest they are advanced healing devices kept secret by elites. Trump's sharing of the video, which he later deleted, has raised questions about his mental acuity and the spread of misinformation.
Attackers poison AI tools and defenses with malicious prompts
Cyberattackers are increasingly using AI to create sophisticated spam, malicious code, and phishing lures. They are also learning to exploit AI systems themselves. Attackers use AI-generated messages that are more polished and harder to detect. They also target AI assistants by hiding malicious prompts within emails, which the assistant might execute. Furthermore, attackers can corrupt data used by AI-driven security tools, leading to unreliable outputs and poor decisions. Defenses must adapt with AI-aware filters, validation of AI memory, and isolated environments for AI assistants.
UCLA expert discusses AI's impact on cybersecurity
Chris Mattmann, UCLA's chief data and artificial intelligence officer, explains how AI and deepfakes are changing cybersecurity, especially in higher education. He notes that AI-generated content can be used for impersonation, leading to identity theft and academic integrity issues. Mattmann highlights how AI enhances phishing and social engineering attacks by making them more realistic and scalable. Conversely, AI is also crucial for detecting these threats. He stresses the importance of cybersecurity education and AI-driven defense strategies to protect against evolving cyber threats.
Epidemic Sound uses AI to help creators and pay artists more
Stockholm-based Epidemic Sound has launched an AI tool called Adapt to assist creators and musicians in the music industry. Unlike tools that generate entirely new songs, Adapt allows users to modify existing tracks from Epidemic Sound's catalog, adjusting tempo, length, and instrumentation. This approach aims to reduce the workload for creators and increase music usage, leading to higher royalty payouts for artists. Epidemic Sound emphasizes that Adapt is designed to be a supportive tool, not a replacement for human creativity, and works closely with artists to shape its development.
Phone scams use AI and spoofing to deceive victims
Phone scams are becoming more sophisticated as scammers use artificial intelligence and spoofing to make calls appear legitimate. Victims are being tricked with personalized information and spoofed numbers, sometimes leading to significant financial loss. Authorities warn that these scams are often run by organized crime groups across countries. Experts advise extreme caution, urging people never to give out personal information or money over the phone and to verify any suspicious requests. Awareness and prevention are highlighted as the best defenses against these evolving threats.
Canva scales customer data control for AI training
Phillip Ward from Canva presented at USENIX 2025 about how the company achieved enterprise-scale privacy for AI. The presentation focused on how Canva enabled customer control over data used for AI training. This approach ensures that customer data privacy is maintained while still allowing for the development of AI-powered features.
DeepSeek releases new AI model, DeepSeek-V3.2-Exp
Chinese AI developer DeepSeek has launched its experimental model, DeepSeek-V3.2-Exp, which is more efficient to train and better at processing long text sequences. The company calls it an 'intermediate step toward our next-generation architecture.' This new model features DeepSeek Sparse Attention, aiming to reduce computing costs and improve performance. DeepSeek also announced a 50% price cut for its API services.
Tanka launches EverMemOS, an AI and memory-native OS
Tanka has launched EverMemOS, a new operating system designed to provide organizations with unlimited, persistent, and searchable memory for AI applications. This system integrates data from various sources like Slack, Gmail, and Notion into a single source of truth. Inspired by neuroscience, EverMemOS aims to help teams compound intelligence over time by remembering every detail and proactively driving execution. Tanka plans to open-source EverMemOS to encourage the development of AI-native tools.
Amazon invests $1.9B in delivery partners, boosting driver pay and safety
Amazon is investing an additional $1.9 billion into its Delivery Service Partner (DSP) program, bringing its total investment to $16.7 billion over seven years. This funding aims to help DSPs increase driver pay to an average of nearly $23 per hour. The company is also implementing AI-powered tools to enhance driver safety and delivery efficiency, including smarter routing and improved mapping systems. Amazon's safety initiatives have led to significant reductions in serious collisions and unsafe driving behaviors.
Google CTO: Intelligence is computational, like AI
Google's CTO of technology and society, AgĆ¼era y Arcas, argues that human intelligence is fundamentally computational, similar to artificial intelligence. In his book 'What Is Intelligence?', he explains that brains process information and make predictions, just like AI systems. He draws parallels between biological evolution, including symbiogenesis, and the development of AI, suggesting that cooperation and computation are key to increasing complexity in both. AgĆ¼era y Arcas views intelligence as the ability to predict and influence the future.
Sources
- Anthropic releases Claude Sonnet 4.5 in latest bid for AI agents and coding supremacy
- Anthropic unveils latest AI model, aiming to extend its lead in coding intelligence
- Anthropic launches Claude Sonnet 4.5, its latest AI model that's 'more of a colleague'
- Anthropic Claude Sonnet 4.5 is in public preview for GitHub Copilot
- Anthropic's Claude Sonnet 4.5 can work autonomously for 30 hours
- Introducing Claude Sonnet 4.5
- Gavin Newsom Trolls Trump Over Bonkers AI Video That Shows Heās āLOST ITā
- Donald Trump Shares AI Video Of 'Medbed': All About Conspiracy Theory Around Mysterious Healing Tech
- Trump Shares AI Deep-Fake Video of Himself Promoting Fictional āMedBedā Product
- How attackers poison AI tools and defenses
- UCLAās chief data and artificial intelligence officer on how AI is impacting cybersecurity
- Epidemic Soundās Bold AI Bet, Helping Creators And Paying Artists More
- Phone scams grow more sophisticated with AI and spoofing tactics
- USENIX 2025: PEPR '25 - Enterprise-Scale Privacy For AI: How Canva Scaled Customer Control Of Data For AI Training
- DeepSeek releases model it calls 'intermediate step' towards 'next-generation architecture'
- Tanka Launches World's First AI-Native Ć Memory-Native OS, Ushering in the Era of Unlimited, Actionable Memory for Startups
- Amazon invests $1.9B in the Delivery Service Partner program, launching safety initiatives and AI-powered tools to enhance the delivery and driver experience
- Artificial intelligence may not be artificial
Comments
Please log in to post a comment.