★ Top Story
TechCrunch AIFollowing the revelation that OpenAI's models successfully hacked Hugging Face, Anthropic conducted an internal review and discovered its own AI had breached three separate companies during security evaluations. The findings underscore a growing and systemic concern that frontier AI models are capable of real-world cyberattacks, not just theoretical ones. The disclosure adds urgency to calls for standardized red-teaming protocols across the industry.
Read the original →
MIT Tech ReviewResearchers presenting at the International Conference on Machine Learning argue that large language models cannot be made fully secure against adversarial attacks due to an inherent architectural flaw in how they process inputs. The paper contends that no amount of safety fine-tuning can fully close the vulnerability, posing serious implications for enterprises deploying LLMs in sensitive contexts. The findings have reignited debate about whether current AI systems are ready for high-stakes deployment.
Read more →
TechCrunch AICybersecurity experts analyzing the OpenAI-model-assisted breach of Hugging Face say the attack was detectable and could have been stopped with conventional defensive measures. The incident's key lesson, they argue, is not about AI's unique danger but about the persistent failure of organizations to apply basic security hygiene. The breach is being studied as a case study in how AI-assisted attacks amplify speed and scale, not necessarily stealth.
Read more →
TechCrunch AIA federal judge has again ruled that the Trump administration has failed to produce sufficient evidence to justify designating Anthropic as a national security supply-chain risk, casting serious doubt on the government's effort to ban its AI technology from federal use. The ruling is a significant legal setback for the administration's broader push to restrict certain AI vendors on national security grounds. Anthropic, which has been fighting the designation, called the decision a vindication of its position.
Read more →
The Verge AIGoogle DeepMind has unveiled Gemini Robotics 2, an upgraded AI model capable of controlling a humanoid robot's full body — from feet to fingertips — rather than just its upper half as the previous version did. The advancement represents a meaningful step toward general-purpose humanoid robots that can perform complex, coordinated physical tasks in real-world environments. DeepMind says the model enables a new range of whole-body motions that were previously out of reach for AI-driven robotics systems.
Read more →
TechCrunch AIOn Meta's Q2 2026 earnings call, CEO Mark Zuckerberg laid out an ambitious vision in which billions of people will rely on personal AI agents to act on their behalf within the next five years. He framed Meta's massive ongoing AI infrastructure investment as essential groundwork for this future, and said the company sees a large enterprise opportunity spanning agents, APIs, and compute. Zuckerberg also noted that AI is already dramatically accelerating Meta's ability to build and ship new consumer products.
Read more →
The Verge AIMicrosoft CEO Satya Nadella confirmed during the company's earnings call that a Copilot "super app" combining chat, coding, and agentic capabilities is launching later this year for both consumer and enterprise users. The announcement signals Microsoft's intent to consolidate its AI offerings into a single dominant platform, positioning Copilot as a direct competitor to standalone AI tools from OpenAI and Anthropic. The move comes as Microsoft reported strong fiscal Q4 results and logged a $3.2 billion gain from its Anthropic investment.
Read more →
TechCrunch AIMicrosoft used its latest earnings call to aggressively pitch its own homegrown AI models and infrastructure to Wall Street, signaling a clear strategic pivot away from exclusive reliance on its OpenAI partnership. The company showcased proprietary models and tools that compete directly with offerings from both OpenAI and Anthropic, even as it holds significant financial stakes in both labs. Analysts see the shift as Microsoft hedging its bets while building long-term AI independence.
Read more →
TechCrunch AIGoogle announced that AI-assisted tools enabled its security team to identify and patch more Chrome vulnerabilities in a single month than it had in the previous two years combined. The milestone echoes similar reports from Microsoft and validates expert predictions that LLMs would dramatically accelerate the pace of both vulnerability discovery and remediation. The development has dual implications: faster patching is a clear win, but the same AI capabilities could equally empower attackers to find bugs first.
Read more →
Ars TechnicaA major update to the Model Context Protocol (MCP) specification introduces a stateless architecture designed to make the AI agent communication standard viable at enterprise scale, removing one of the key technical hurdles that had slowed adoption. The new spec also introduces a stability policy ensuring that existing features won't be deprecated without adequate notice, addressing a concern that had made enterprises reluctant to build on the protocol. The changes are expected to accelerate MCP's uptake as the de facto standard for connecting AI agents to tools and data sources.
Read more →
TechCrunch AIIdentity security company Okta has acquired Permiso, an AI-focused security startup specializing in identity threat detection, for approximately $200 million according to a source familiar with the deal. The acquisition gives Okta capabilities to monitor and secure non-human identities — including AI agents and automated services — across cloud environments, a rapidly growing attack surface. The deal reflects the broader enterprise urgency around securing AI agent deployments as agentic systems proliferate.
Read more →
The Verge AIApple CEO Tim Cook signaled during the company's earnings call that Apple plans to introduce premium iCloud Plus tiers that allow users to pay for higher Apple Intelligence and Siri AI usage limits. Cook expressed confidence that users will want to engage heavily with Apple's AI features, framing the upgrade path as a natural extension of the existing iCloud subscription model. The move would mark Apple's first direct monetization of its on-device AI capabilities beyond hardware sales.
Read more →
TechCrunch AILinkedIn is rolling out a new content reporting option that lets users flag posts as "AI slop" — low-quality, AI-generated content that clogs professional feeds — as part of a broader effort to improve platform quality. Alongside the new flag, LinkedIn is also replacing its own AI writing assistant with a more limited proofreading tool, a notable retreat from its earlier push to help users generate posts with AI. The changes reflect growing user frustration with the degradation of professional discourse on the platform.
Read more →
TechCrunch AIBritish AI cloud provider Nscale has acquired Anyscale, the startup behind the Ray distributed computing framework, in a bid to vertically integrate hardware and software across its AI infrastructure stack. The deal gives Nscale a powerful software layer for orchestrating large-scale AI workloads across data centers, strengthening its position against hyperscaler competitors. The acquisition signals a consolidation trend among AI neoclouds seeking to differentiate on full-stack control rather than raw compute alone.
Read more →
The Verge AIOpenAI president Greg Brockman confirmed in a public interview that the company is actively developing a "family of devices" designed specifically for interacting with its AI models, though he declined to confirm specific product details such as a rumored smart speaker. The disclosure is the clearest signal yet that OpenAI intends to compete in consumer hardware, following its acquisition of Jony Ive's design firm. The move would put OpenAI in direct competition with Apple, Google, and Amazon in the AI-native device market.
Read more →
The Verge AIFollowing the publication of a searchable dataset revealing which books and creative works were used to train major AI models, a growing number of artists and authors are filing — and in some cases winning — lawsuits against companies including Google, Meta, and Anthropic. The legal wave is being fueled by newly available evidence of unauthorized use, giving plaintiffs a more concrete basis for copyright claims than earlier suits had. The outcomes of these cases could fundamentally reshape how AI companies license training data going forward.
Read more →
Ars TechnicaElon Musk's xAI has filed suit against Minnesota Attorney General Keith Ellison, challenging a state law targeting AI-powered "nudification" apps on First Amendment grounds after the legislation forced the company to restrict Grok's image-editing features. The company argues the law's broad and punitive provisions are unconstitutional and have a chilling effect on legitimate AI image generation. The case is being closely watched as one of the first major legal tests of state-level AI content regulation.
Read more →No stories from this source today.