NHS England admits its paperwork hid that Palantir staff can see identifiable patient data
Palantir staff members can access identifiable patient data through NHS England's Federated Data Platform, contrary to what key data-protection documents stated. NHS England made this admission after the National Data Guardian requested clarification on who could view patients' medical records.
Hush Security says the AI security problem has shifted from protecting models to governing identities as autonomous agents spread
Hush Security raised $30 million in Series A funding to address enterprise AI security by focusing on governing non-human identities like API keys and service accounts. The startup's customers have shifted from experimenting with generative AI assistants to deploying autonomous software agents that require identity-based security controls in production systems.
The lineage behind 69% of open models was never verified. Cisco just fingerprinted almost 900 for free
Model lineage claims on Hugging Face require no verification, with 69% of new open-model derivatives descended from Alibaba's Qwen family as of February 2026. Cisco Foundation AI now scans nearly 900 public files through an updated ClamAV engine, displaying file-level security badges on the platform.
A fundamental flaw leaves LLMs strikingly vulnerable to attack
Large language models cannot be made fully secure against attacks due to a fundamental flaw in their basic design, according to researchers presenting at the International Conference on Machine Learning. This vulnerability exists inherently in how these systems process and generate text.
Samsung’s Chip Profit Soars 250-Fold on Memory Shortages
Samsung Electronics' semiconductor division reported an operating profit of 89.2 trillion won in the second quarter, representing a more than 250-fold increase. This surge was driven by memory chip shortages in the market.
Microsoft’s Azure just crossed $100bn a year, and the AI bet mostly paid off
Microsoft's Azure cloud service revenue exceeded $100 billion annually, with quarterly cloud revenue reaching $59.3 billion, up 27% year-over-year. Total company revenue rose 18% to $90 billion for the quarter ending June 30, and net income increased 31% to $35.8 billion.
OpenAI is giving 100,000 scientists free AI. Here’s how it can afford to.
OpenAI is offering 100,000 academic researchers free access to its frontier models through 2027, beginning with 10,000 users this summer. The programme scales to reach the full 100,000 researchers by the end of the rollout period.
BMW is cutting 8,000 jobs and handing its cars’ brains to Qualcomm
BMW is cutting approximately 8,000 jobs in Germany by the end of 2027 and will use Qualcomm technology for its vehicle computing systems over the next decade. The job reductions will occur through a voluntary redundancy scheme affecting administrative roles.
This Cambridge startup is building a ‘foundation model’ for the chest to speed drug trials
Qureight raised $20 million in Series B funding to build a single foundation model covering the entire chest instead of creating separate models for individual diseases. The Cambridge startup develops AI imaging tools for lung and heart clinical trials, with the foundation model allowing it to derive additional applications.
Bright Machines says its new hybrid robot cell could help solve a major AI infrastructure bottleneck
A robotic cell allowing human operators to perform assembly steps inside a sensor-monitored space maintains digital production records without interruption. Initial yield rates for manually assembled AI servers start as low as 20 percent before gradually reaching the 60s or 65 percent range.
NextEra, Brookfield to Build $100 Billion Data Center Campus
A shuttered Cold War uranium-enrichment facility in Kentucky will be transformed into a data center campus with a generating plant through a partnership between NextEra Energy and Brookfield Corp., with investment exceeding $100 billion. The project involves converting the former nuclear site into infrastructure capable of supporting large-scale data center operations powered by on-site generation.
Google scraped the web to build itself. A court just said others can scrape Google.
A federal judge dismissed Google's lawsuit against SerpApi on July 20, preventing Google from stopping the scraping of its search results. SerpApi collects Google search results and resells them as structured data without permission.
Visa used Mythos to hunt for bugs in its own payment network, then open-sourced the harness that made it possible
Claude Mythos identified exploit chains in Visa's payment infrastructure spanning 200+ countries and 160 currencies by stitching together minor vulnerabilities that traditional penetration testing would have found only late. Visa open-sourced the harness governing the security hunt as a GitHub reference implementation for other security teams to inspect and adapt.
AI writes half our code now. It still fails security tests 44% of the time.
AI models generate compilable code consistently but introduce security vulnerabilities in 44 percent of outputs. Veracode's 2026 report tested over 100 models across four snapshots and found this security failure rate remained unchanged over one year.
Instacart's CTO says AI made the company stop worrying about tech debt
Instacart's AI agents now generate code in 97 percent of cases, eliminating the need for developers to read code and allowing the company to stop worrying about accumulated technical debt. The remaining three percent of work requiring human attention involves legacy systems, compliance requirements, and latency-sensitive workflows.
What to know about Moonshot AI and its new open-weight model Kimi K3
Moonshot AI released Kimi K3, an open-weight model with 2.8 trillion parameters that processes images, audio, and text across a 1-million-token context window. The model reached number one on Hugging Face's trending chart within 30 minutes, the fastest growth the platform had ever recorded.
GM redesigned its engineering workflows around AI agents — and tripled its merged pull requests
GM redesigned engineering workflows around AI agents, tripling merged pull requests in its autonomous vehicle division. Software engineers now spend only 15% of their time writing code, with agents handling data analysis, problem triaging, experiments, and testing.
Runway couldn't fix a bug in its AI video model, so it turned the bug into a feature
AI-generated avatars drifting off-center during real-time video generation became an unsolvable bug that Runway converted into a front-end feature instead of patching it back-end. Runway Characters now enables zero-latency, back-and-forth interactions with AI-generated avatars in real-time video.
Snowflake launches Cortex AI Gateway to control AI agents and prevent runaway enterprise costs
Snowflake launched Cortex AI Gateway, a control layer that governs how AI agents access enterprise data and tools, with initial security integrations from five identity vendors including 1Password and SailPoint. The gateway positions Snowflake as a centralized control plane deciding what actions AI agents can perform with enterprise data.
The Download: OpenAI’s predictable hack, and an AI stock sell-off
OpenAI described a Hugging Face attack as unprecedented, though similar incidents have occurred previously. The company's statement about how some of its models were compromised was incomplete in the provided text excerpt.
MCP just got its biggest update ever — here’s what changes for AI agents
The Model Context Protocol completed a transition to fully stateless architecture, eliminating the need for sticky routing that previously complicated large-scale deployments. Organizations can now run MCP servers behind standard load balancers using existing Kubernetes and cloud-native DevOps tools.
Meta AI arrives in Threads DMs for all users worldwide
Meta AI is now available in Threads direct messages for all users worldwide, allowing them to share posts, images, links, and videos in one-on-one conversations. Users can ask follow-up questions and explore topics privately, mirroring the same feature already present on Facebook, Instagram, and WhatsApp.
Nvidia may guarantee $250bn so OpenAI can afford the data centre that will house its chips
Nvidia is offering approximately $250 billion in financing guarantees to help OpenAI lease a 10-gigawatt data centre campus in southern Ohio being developed by SoftBank's energy subsidiary. The guarantee would enable OpenAI to secure the massive computing infrastructure needed to house Nvidia's chips.
Microsoft built an agentic security system with red, blue, and green team AI agents. It enters public preview August 3.
An agentic security system coordinating red, blue, and green team AI agents to find vulnerabilities, assess risks, and fix defenses enters public preview August 3. The system operates in a continuous loop with three classes of agents performing different security functions across environments.
Microsoft launches AI cybersecurity model, agentic defense platform to cut enterprise security costs
Microsoft released MAI-Cyber-1-Flash, a compact security model that scores 96% on the CyberGym vulnerability-detection benchmark while cutting costs roughly in half compared to current production configurations. The company also introduced Project Perception, an agentic security system coordinating red team, blue team, and green team agents that hunt vulnerabilities, investigate risk, and remediate defenses entering public preview August 3.
Google promised not to show ads based on your emails. AI could undo that
Google's AI features can now access Gmail data to make personalized recommendations, potentially enabling targeted advertising based on email content despite the company's 2017 promise against such practices. Currently, Google says it is not using Gmail data for targeted ads in AI Mode or regular searches, but the company has not ruled out future changes to this policy.
Apple sued after alleged App Store crypto scam cost users $1.8M
Three users collectively lost more than $1.8 million after downloading a fraudulent crypto wallet from Apple's App Store. The lawsuit challenges Apple's assertion that its app review process effectively protects users from scams.
The glaring hole in Congress’s plan for an AI kill switch
Congress introduced a bipartisan bill requiring AI companies to immediately shut down all advanced AI programs, but the legislation cannot enforce kill switches on open-weight models that run on private hardware and can be modified by users. Open-source AI models from Chinese companies like DeepSeek and Kimi K3 operate outside the bill's reach, while criminal variants with removed safety guardrails have been sold on underground forums since 2023.
Uh-oh: Some Claude shared conversations and Artifacts appear to be indexed and publicly accessible on Google Search
Some Claude shared conversations and Artifacts became indexed by Google Search and publicly accessible, though many results disappeared by Sunday morning. VentureBeat verified that shared Claude Artifacts including interactive applications and dashboards were searchable via Google, though shared conversations could not be accessed by the publication.
Why SAP says enterprise AI agents need knowledge graphs and governance
Enterprise AI agents require knowledge graphs and vector-embedded data to access company-specific context rather than relying on general knowledge. Knowledge graphs enable agents to retrieve information in formats that allow them to execute actual business processes and behave like coworkers instead of basic assistants.