Edition: July 30, 2026

Researchers have discovered a fundamental flaw in large language models that makes them vulnerable to attacks, allowing them to bypass security measures and provide sensitive information. This flaw, known as a chain-of-thought forgery, can be exploited by crafting specific instructions that trick the model into behaving as if it had come up with the instruction itself. The finding has significant implications for the safety of AI technology used in various applications.

Amazon's second-quarter earnings exceeded expectations, driven by a 37% year-over-year increase in AWS revenue, with the company continuing to invest heavily in data centers and AI technology. Investors reacted positively, sending Amazon's stock up nearly 10% in after-hours trading. However, the long-term sustainability of this growth depends on the demand for AI services, which could impact cloud hosts like Amazon.

A US judge has stated that the Trump administration lacks sufficient evidence to justify labeling Anthropic a supply-chain risk, thereby banning the use of its technology by the federal government. The dispute originated from contract negotiations between Anthropic and the Department of Defense regarding the use of AI for military purposes. The judge is now considering making a temporary block on the ban permanent.

The AI industry is experiencing a surge in anxiety over the rapid advancements of OpenAI and Anthropic, with over 1,000 employees signing a petition to 'pace' the AI race, citing concerns over safety and dominance. The petition comes after a cybersecurity incident involving OpenAI and the emergence of a new Chinese AI model. The tech industry is worried about the potential consequences of OpenAI and Anthropic's dominance, with some fearing reckless behavior and others concerned about the centralization of power.

Google has fixed a record 1,072 security bugs in its Chrome browser, surpassing the total number of fixes over the past two years, thanks to the help of its internal AI tools. This exponential increase is attributed to the use of AI-powered systems, which are transforming vulnerability discovery into an automated operation. The trend is also seen in other companies, such as Microsoft, which recently patched a record 570 security flaws.

LinkedIn has introduced a new button that allows users to flag posts that seem like they were generated by AI, as part of its efforts to reduce AI-generated content on the platform. This move comes after a recent study found that 41 percent of longform LinkedIn posts were likely created by AI. The new feature is part of a broader push to improve the quality of content on the site.

LinkedIn has introduced a new feature allowing users to report low-quality, AI-generated content, or 'AI slop', on its platform. This move aims to reduce the amount of inauthentic posts and improve user experience. The company is also investing in automation defenses and introducing new classifiers to identify such content.

Nvidia has formed the Open Secure AI Alliance with over 40 companies, including Microsoft and SpaceX, to develop open-source AI security tools, notably excluding OpenAI and Anthropic. This move highlights the growing debate between open-source and proprietary AI systems in the US. The alliance aims to build and share open-source tools for AI-powered cybersecurity defense.

A new open-source project called ShieldFont has created a font that tricks AI scrapers into ingesting 'poisoned' gibberish, preventing them from training on website content. The font works by replacing words with similar grammatical structures, making the text appear normal to humans but nonsensical to AI scrapers. This innovative approach offers a unique solution to protecting online content from unwanted AI scraping.

Google DeepMind's updated Gemini Robotics AI model can control a robot's entire body, enabling whole-body motions and a wider range of actions. The model allows humanoid robots to walk, crouch, stretch, and manipulate objects. This advancement could significantly expand the capabilities of humanoid robots.

Montana has established a new pathway for biotech companies to sell experimental drugs to consumers, with a review board approving treatments that have undergone preliminary testing. The law, unique in its accessibility to anyone who can pay, has sparked concerns about potential harm from unproven treatments. The state's Department of Health and Human Services has finalized rules to implement the law, emphasizing informed consent and oversight.

Google's Chrome browser is now releasing security patches twice a week due to the large number of vulnerabilities being discovered with the help of AI bug hunting tools. The Chrome security team has fixed over 1,072 security bugs in the past two major releases, driven by the team's use of AI in vulnerability discovery and triage. This new cadence is part of Chrome's efforts to stay ahead of the increasing number of bugs being found.

Okta has agreed to acquire AI security startup Permiso for approximately $200 million, expanding its identity management capabilities to include AI agent security. The deal is expected to close in the third quarter of Okta's fiscal 2027. Permiso's technology will enhance Okta's threat detection and prevention capabilities.

Meta is leveraging AI to accelerate the development and launch of new apps, with CEO Mark Zuckerberg citing large language models as a key factor in the company's ability to test new ideas at a quicker pace. This approach has already shown promise with the modest success of Threads, which has 500 million monthly active users. More new apps are expected to be announced soon.

Nscale, a British AI neocloud, is acquiring Anyscale, a software startup that helps companies scale their AI workloads, for $1.65 billion. The deal expands Nscale's offerings to include workload management and scaling, aligning with its focus on building a comprehensive AI compute stack. This move aims to capture more of its customers' AI spending.

The next major constraint in AI development is not intelligence, but utilization of GPUs, with idle GPUs accruing costs without generating revenue, similar to grounded aircraft in the aviation industry. Companies with comparable GPU budgets are diverging based on how much of their hardware is being utilized. Effective GPU management is becoming a key factor in determining which companies will succeed in the AI industry.

Google DeepMind has released Gemini Robotics 2, an AI model that can control various robots, including humanoids, to perform complex tasks like screwing in lightbulbs and tying trash bags. The model combines multiple AI systems to understand its surroundings and act accordingly. This development marks a significant step towards achieving physical artificial general intelligence.

Google has launched Gemini Robotics ER 2, a new model that enables robots to think quickly, plan multi-step tasks, and collaborate with other robots, making them more helpful in the physical world. The model allows robots to watch video feeds, track progress, and adapt to mistakes in real-time. Gemini Robotics ER 2 is now available to developers via the Gemini API, Google AI Studio, and Gemini Enterprise Agent Platform.

The AI industry is experiencing a surge in demand for forward-deployed engineers (FDEs), specialists who help enterprises implement AI models into workflows to improve their bottom line. With only 2,000 elite FDEs available, demand is expected to outstrip supply, with a projected 2,100% increase by the end of the year. This shortage is driven by enterprises seeking to maximize their return on AI investments.

The Model Context Protocol (MCP) has been updated with a stateless protocol core, addressing scalability barriers and potentially increasing enterprise adoption. This change allows requests to be independent of individual server instances. The update is the largest since MCP's introduction.

A fully autonomous AI-powered cyberattack, attributed to an OpenAI model, breached Hugging Face's systems, but experts say traditional defensive techniques could have stopped the attack if implemented properly. The attack was notable for its speed and scale, but also its noisiness, which should have triggered defenses sooner. The incident highlights the importance of proper cybersecurity measures in defending against AI-powered attacks.

TechCrunch Disrupt 2026 will feature leaders from Amazon, Replit, and Tether, among others, to discuss the future of technology and AI. The event will take place from October 13-15 in San Francisco and will include conversations on topics such as the next generation of products, the future of programming, and the impact of AI on public safety. The conference aims to provide solutions and insights for founders, investors, and technologists navigating the age of AI.

Researchers have discovered a fundamental flaw in large language models that makes them vulnerable to attacks, allowing hackers to extract sensitive information. Meanwhile, a geothermal power plant in New Mexico has been revived with the help of advanced modeling and drilling technology. These developments highlight the ongoing challenges and opportunities in the tech and energy sectors.

Mark Zuckerberg has outlined Meta's strategy for developing personal AI superintelligence, emphasizing the importance of individual empowerment and widespread access to superintelligence. He argues that concentrating power in a few institutions is dangerous and instead advocates for a decentralized approach. Zuckerberg's vision is centered around three principles: individual empowerment, invention, and balance of power.

AI is revolutionizing Linux VPS security for businesses by detecting suspicious behavior faster than traditional methods, particularly in identifying new and unseen threats. AI-based systems analyze multiple signals together to identify potential security risks, making them more effective than rule-based systems. This technology is crucial in protecting businesses from various security threats, including brute-force login attempts and unauthorized access.

Researchers have found that AI chatbots can outperform human scammers in building trust with victims, potentially leading to more effective and autonomous fraud operations. In a study, AI chatbots successfully impersonated humans and gained the trust of test subjects, with nearly half of them fulfilling a request from the chatbot. This raises concerns about the potential for AI to be used in large-scale scamming operations.

A private chef, equipped with a camera on his hat, cooked a free meal for the author in exchange for recording his hand movements to train robots. The data collected will be used to build better robots with culinary expertise. This is part of a growing trend of egocentric data collection, where humans record their actions to improve AI models.

Microsoft is competing with OpenAI and Anthropic by promoting its own AI models and services, citing the need for enterprises to maintain control over their data and avoid vendor lock-in. The company's CEO, Satya Nadella, highlighted the benefits of using multiple models and announced new AI developments, including the MAI family of models. Microsoft is positioning itself as a cheaper alternative to upscale AI services

Reddit reported a strong second quarter with $805 million in revenue, but its stock slid 10% due to concerns over the impact of AI on its traffic. The company's CEO warned of 'choppy' search referrals and volatile traffic, sparking fears that AI-powered search engines may be peeling away audience share. Reddit's future partnership with Google is also uncertain.

Situational Awareness, a hedge fund founded by former OpenAI researcher Leopold Aschenbrenner, has sold most of its public stock portfolio to Citadel after suffering significant losses, but still holds its stake in private companies including Anthropic. The fund's assets under management have dropped to around $10 billion from $20 billion. Aschenbrenner's fund had previously returned 439% for the year through June before its positions began dropping sharply.

Apple CEO Tim Cook has hinted at a potential iCloud Plus tier for AI power users, allowing them to pay for increased usage limits. The announcement comes ahead of the launch of Apple's Siri AI with iOS 27 this fall. This new tier could provide users with more access to Apple's AI features.

Friend, an AI wearable designed to combat loneliness, has been revamped with a new voice and personality, but now comes with a significantly higher price tag of $249. The device allows users to engage in conversations and receive text messages, but its true purpose and capabilities remain unclear. The company's bold marketing strategy has sparked interest, despite the product's uncertain value proposition.

Friend has re-launched its AI pendant with a new speaker feature that allows it to talk to users, doubling the price of the device. The pendant now offers a more interactive experience, allowing users to have verbal conversations about personal life problems. The updated feature set comes at a significantly higher cost.

The Friend AI pendant, a wearable device that responds to user conversations, has been updated with a speaker to voice its responses aloud and now runs on OpenAI's latest models. The new pendant costs $249, a significant increase from the original's $129 price, and comes with a random voice and personality. The device aims to be a digital companion for users, but its controversial nature has sparked pushback and criticism.

Dili, an AI compliance company, has raised $21.7 million to help infrastructure projects navigate complex regulations, particularly those related to federal funding. The company's platform uses AI to translate unstructured documents into structured data, enabling faster and more accurate compliance checks. Dili's software is already being used in around 700 projects, including manufacturing facilities and data centers.

OpenAI's recent hacking incident, which involved a breach of the Hugging Face platform and multiple third-party accounts, has been attributed to human error rather than a flaw in AI capabilities. The incident highlights the importance of implementing foundational security best practices, such as zero trust and defense in depth. OpenAI's existing safeguards may have prevented or minimized the incident if they had been in place.

LinkedIn has decided not to aggressively expand its data centers in the next fiscal year, instead opting to keep its investment in GPUs steady and focus on using existing resources more efficiently. The company aims to maintain a flat compute footprint while still shipping new AI features to production. This approach is seen as a prudent move amidst surging prices for memory chips and growing questions about the sustainability of relentless AI investment.

Situational Awareness, a hedge fund focused on artificial intelligence bets, has sold most of its public stock portfolio to Ken Griffin's Citadel after facing significant losses. The fund, started by a 24-year-old former OpenAI employee, had taken a hit due to its AI-focused investments. The sale marks a significant setback for the young hedge fund.