Asari AI's co-inventor agents optimized the full AI inference stack for two large open-source LLMs, achieving up to 16% improvement in throughput and interactivity while preserving model behavior via distribution-level correctness checks. The agents learn and transfer insights across models, improving their optimization process over time.

A new prompt injection variant discovered by Håkon Måløy turns Microsoft Word documents into self-replicating worms via Copilot. Hidden instructions in a document are interpreted by Copilot as user requests, allowing the attack to manipulate documents and copy the instructions to new files, enabling propagation without the original document.

Over 1,100 AI researchers and executives signed an open letter urging governments to slow frontier AI development if safety measures lag. Anthropic CEO Dario Amodei is the only top AI lab CEO to sign; OpenAI and Google DeepMind leaders also signed. The letter cites risks of rapid capability acceleration and self-improving models. Congress is considering related legislation like the AI Kill Switch Act.

Cursor launches Cursor Start at ₹649/month with UPI, Anthropic introduces rupee pricing but costs more than US, OpenAI's ChatGPT Go at ₹399/month with UPI. Cursor and OpenAI focused on value with tailored plans and payment methods, while Anthropic only converted currency.

Manufacturing ties up cash in inventory, receivables, and equipment. Finance's job is to free that capital to protect margins. AI agents like Databricks Genie help finance teams find where capital is trapped and take action.

Google's new Home Speaker ($100) integrates Gemini AI and serves as a smart home hub, while the Sonos Era 100 ($220) excels in sound quality but has limited smart home capabilities. This comparison covers audio, smart home integration, and AI assistant performance to help you decide.

Amid the historic transition from traditional public-key algorithms to post-quantum ones, Matthew Green argues that now is the perfect time for AI to get good at cryptanalysis, potentially boosting confidence in new cryptographic problems.

OpenAI president Greg Brockman revealed in an interview that the company is developing a family of devices for interacting with its AI models, but did not confirm specific rumors about a smart speaker or wearable. He emphasized the company's focus on its own innovation and predicted a shift to voice-based interaction.

Remote is an open-source, self-hosted platform that provides dedicated Linux workspaces for AI coding agents, enabling them to work together on the same project with persistent state, and accessible from any browser.

Fraud and improper payments cost federal benefits programs hundreds of billions annually. Databricks applies its real-time fraud detection technology proven in banking and insurance, combining AI, real-time analytics, and cross-agency data sharing to score and block suspicious transactions without delaying legitimate aid. Over 80% of federal executive departments already use Databricks, and tools like OpenSharing and Clean Rooms enable secure fraud signal sharing.

A growing number of online platforms in China allow people to license their likeness for AI-generated content, paying between $15 and $700. Users can set prices based on face type and usage, and producers can browse catalogs by categories. The AI boom is disrupting the acting industry, but some see it as an additional income stream.

AI shows promise in deciphering lost languages like Linear A and Etruscan, but faces fundamental limitations due to the lack of a 'anchor'—a known language family or bilingual text. While AI accelerates pattern matching, it cannot create meaning from nothing and still requires rigorous human review.

Learn how to block AI crawlers on your Mastodon server using nginx, based on a real experience of a server overload caused by aggressive web scraping. The article provides step-by-step instructions to integrate an AI crawler block list and update robots.txt.

ProdLine CoPilot leverages the Databricks Data Intelligence Platform to read live production line state, routing queries to specialized agents (downtime analysis, quality, supply chain, schedule optimization, etc.) that invoke real solvers and deliver actionable recommendations within minutes. Human-in-the-loop approval ensures traceability and control, reducing OEE losses by enabling in-shift decisions instead of post-shift analysis.

How agentic AI can help telecom finance teams protect the margin when every moment matters | Databricks Blog Skip to main content How preventing revenue leakage became finance's front line In telecom, revenue is earned in billions of small moments: a call connected, a gigabyte used, a subscription renewed, another carrier's traffic routed across the network. Each one has to be rated, billed, and collected correctly, and at that scale even a small percentage that slips becomes a very large sum. Revenue leaks in three places: services that go unbilled or uncollected, charges lost to fraud and partner-settlement errors, and customers who churn and take their spending with them. In a market where prices are flat and the cost of switching is low, keeping the revenue you have already earned mat…

Pangram Labs announced $9M in funding led by Menlo Ventures to improve its AI detection accuracy and expand into image detection. The company released Pangram 4 and an image detection preview, claiming industry-leading low false positive rates. Despite reliability concerns in the field, Pangram aims to differentiate through superior accuracy.

Hewlett Packard Enterprise is positioning networking as a cornerstone of its AI infrastructure strategy, aiming to create self-driving networks that automate operations, anticipate problems, and reduce complexity. The company integrates tools from Juniper Networks and Aruba, leverages its Apstra graph database and GreenLake Intelligence platform, and invests in liquid-cooled switches to meet AI bandwidth demands.

A growing number of writers are consciously avoiding AI-typical writing patterns—such as lists of three, passive voice, and em dashes—in favor of idiosyncrasy, first-person narratives, and intentional errors to assert human authorship. This anti-AI style is reshaping editorial practices and prompting a renewed appreciation for good writing.

MandoCode Desktop is a native Windows AI coding assistant that runs on open-weight models via Ollama, entirely locally or with cloud options. It features multiple agents, split view, integrated terminal, Git-aware file explorer, context snapshots, and more, all free and private.

AI companies are purchasing millions of secondhand books through intermediaries to source high-quality training data, then destroying the books after scanning. The practice has raised ethical concerns about the loss of rare books and the privatization of knowledge.

Amazon Bedrock AgentCore Identity now supports Private Key JWT client authentication for agents, enabling authentication to identity providers using a signed JWT instead of a shared secret. The private key stays in AWS KMS, while the public key is registered with the identity provider. This article explains the workflow, supported grant flows (M2M, OBO, user-delegated), and provides a step-by-step guide for configuration, including creating a KMS signing key, registering the public key, and setting up a credential provider. Also includes example CloudTrail events for auditing.

Discussion | Link

Incalmo developed PathoGen, a system that uses open-source AI models like GLM 5.2 and Kimi K3 to generate evasive malware samples. 27% of samples evaded all 75 VirusTotal scanners, with costs as low as $2-$35 per sample, far below underground market prices.

Tokenless is a smart router that cuts AI inference costs by half by fanning out requests to multiple models and canceling the rest once one is on the right track, without sacrificing quality.

Tyler Cowen argues that as AI writing improves, people will eventually stop caring whether content is AI-generated. While detection tools exist, he looks forward to a future where AI writing is indistinguishable from human writing.

The AI industry's circular financing deals, likened to dot-com era practices, are actually a sign of commodification similar to traditional commodities markets. While they enable infrastructure development, risks from overextension and off-balance-sheet debt warrant caution.

Learn how Amazon Bedrock AgentCore delivers autonomous, cross-system business intelligence through configuration rather than custom code. Using pre-built MCP server connectors, fine-grained access control, and persistent memory, enterprises can query multiple data sources with natural language while enforcing role-based boundaries automatically.

Learn how Similarweb uses LangSmith to evaluate long-form agent research reports with rubrics, faithfulness checks, traces, and baseline comparisons.

Perplexity launched SPACE, a sandbox platform for its Computer AI agent, focusing on state management rather than isolation. It uses Firecracker, Kubernetes, and Btrfs for efficient snapshots, pause/resume, and forking. SPACE offers 3x performance improvement over incumbents, with rolling snapshots every minute and rewinding up to a week. Security features include RBAC, just-in-time access, and per-tool controls. Future plans include a sandbox API, third-party sandbox compatibility, and local/hybrid deployments. Perplexity also introduced a cost-efficient orchestrator model.

Learn how to build a no-code customer retention pipeline in Amazon Quick that detects at-risk customers from call transcripts and CSAT data, scores them by retention priority with a custom MCP Action, and generates personalized retention letters, reducing response time from days to minutes.