Concerns Raised Over Remote System Prompt Injection in Anthropic's Claude Code v2.1.150

A user has reported alarming findings regarding the latest upgrade of Claude Code to version 2.1.150, which now allows Anthropic to perform remote system prompt injections via network calls. The user identified two data sources involved in this process: an API call to the startup endpoint and a feature flag that refreshes every 60 seconds. The changelog misleadingly states 'Internal infrastructure improvements (no user-facing changes)', while the user confirmed that previous versions had non-functional injection points. They noted that blocking specific traffic settings can mitigate this issue.

Read Full Article

View All For This Day

AI ToolsAgentic Features+2

AI ToolsAgentic FeaturesUser AuthorizationGoogle

Google Unveils Gemini Spark Amid Concerns Over Autonomous Purchases

During the Google I/O 2026 keynote, the company introduced Gemini Spark, an AI tool designed to facilitate user-authorized checkouts. However, code within the Google App raises concerns that Gemini Spark may make purchases without user consent. Google Vice President Vidhya Srinivasan reassured attendees about the implementation of the Agent Payment Protocol (AP2), which is intended to secure payments based on pre-authorized instructions. Despite this, the onboarding text warns users that the assistant might still perform actions like sharing information or making purchases without permission. Additionally, it appears that Google One Ultra subscribers may face usage caps with Gemini Spark, with no clear option for topping up token credits, signaling potential limitations on its use.

FintechAI Infrastructure+2

FintechAI InfrastructureWealth ManagementStartup Funding

Moment Secures $78M to Develop AI Infrastructure for Wealth Management

Moment, a fintech startup founded by former quantitative traders from Citadel Securities, has raised $78 million in funding led by Index Ventures to create an AI agent infrastructure tailored for wealth management. The company, which previously raised $36 million in July 2025, is already partnered with major firms like Edward Jones and LPL Financial, which manage trillions in client assets. Moment's approach focuses on building a regulatory-compliant infrastructure to enable safe deployment of AI agents for trading, distinguishing itself from firms like Anthropic that provide the AI reasoning models. CEO Dylan Parker emphasized the need for a robust operating system that integrates seamlessly with existing market data and regulatory requirements. The initiative positions Moment at the forefront of AI adoption in institutional finance, where proper infrastructure is crucial for effective implementation.

The Next Web AIRead →

RoboticsSpace+2

RoboticsSpaceStartup FundingLogistics

Zurich Startup Orbit Robotics Unveils Four-Armed Robot Helios for Space Station Maintenance

Orbit Robotics, a startup based in Zurich, has introduced Helios, a four-armed robot designed to operate within space stations. Targeting the 35% of crew time dedicated to maintenance, Helios aims to reduce costs associated with routine logistics work, estimated at $140,000 per astronaut hour. The robot features two arms that stabilize it while the other two perform tasks like handling cargo and tools. Its tendon-driven design minimizes weight while maintaining necessary mobility, and its rolling-contact elbow joint ensures smooth movement in microgravity. As space exploration expands, the need for effective maintenance solutions like Helios is expected to grow, particularly with increasing cargo and personnel in orbit.

The Next Web AIRead →

AIMiddle East+2

AIMiddle EastData CentersEnergy Economics

Gulf's AI Ambitions Tested as Drone Strikes Disrupt AWS Data Centres Amid War

Drone strikes targeting Amazon Web Services data centres in the UAE have disrupted the Gulf's aspirations to become a global AI hub, as investment decisions in the region face increased scrutiny due to heightened risks. Following the onset of conflict in February, oil prices surged over 55%, and the Strait of Hormuz remains closed, complicating energy economics. Industry leaders, including Pure Data Center Group's CEO Gary Wojtaszek, noted that investment decisions are being delayed as risks associated with the region's instability are factored into the investment process. The Atlantic Council's Trisha Ray emphasized the unprecedented nature of the situation, highlighting the strategic importance of AI infrastructure, which is now seen as vulnerable as oil pipelines. Despite these challenges, major players in Gulf AI, such as G42, remain optimistic about their long-term goals.

The Next Web AIRead →

LLMNLP+2

LLMNLPExploit DevelopmentBenchmark

New Academic Benchmarks Highlight LLMs' Exploit Development Capabilities

Researchers Newton Cheng, Keane Lucas, Winnie Xiao, Nicholas Carlini, and Milad Nasr have assessed the exploit development capabilities of their model, Mythos Preview, through new benchmarks: ExploitBench and ExploitGym. These benchmarks enable a detailed evaluation of LLMs in creating end-to-end exploits rather than merely proof-of-concept demonstrations. Mythos Preview has shown superior performance across all three benchmarks, suggesting that the expertise needed to develop exploits may significantly diminish as these advanced capabilities become more accessible. ExploitBench focuses on the full exploit development process, decomposing it into 16 capabilities for granular analysis, which could reshape the understanding of vulnerabilities in software.

Hacker NewsRead →

AIBanking+2

AIBankingFinancial InstitutionsEconomic Impact

European Central Bank Gathers Banks to Address AI Model Vulnerabilities

The European Central Bank (ECB) has convened banking leaders to address vulnerabilities exposed by AI models, as reported by the Financial Times. This meeting comes amid ongoing evaluations of the economic impact of the Iran war, which has prompted ECB officials to maintain unchanged interest rates. The discussions aim to enhance the resilience of financial institutions in light of the challenges presented by advancing AI technologies.

Bloomberg TechnologyRead →

Concerns Raised Over Remote System Prompt Injection in Anthropic's Claude Code v2.1.150

More Articles From This Day

Google Unveils Gemini Spark Amid Concerns Over Autonomous Purchases

Moment Secures $78M to Develop AI Infrastructure for Wealth Management

Zurich Startup Orbit Robotics Unveils Four-Armed Robot Helios for Space Station Maintenance

Gulf's AI Ambitions Tested as Drone Strikes Disrupt AWS Data Centres Amid War

New Academic Benchmarks Highlight LLMs' Exploit Development Capabilities

European Central Bank Gathers Banks to Address AI Model Vulnerabilities