IT News Roundup: AI Agent Exploits, Ransomware Frameworks, and Custom Silicon - July 6, 2026

The past week has brought significant developments across AI security, malware threats, custom silicon, autonomous vehicles, and cloud infrastructure. From a critical exploit chain that turns AI browsing agents into remote code execution vectors, to a modular malware framework embedding ransomware specifically designed to target recovery systems, the threat landscape continues to evolve rapidly. Meanwhile, major players in AI and cloud infrastructure are making strategic moves that will shape the industry for years to come.

AutoJack Exploit Chain Turns AI Browsing Agents Into Remote Code Execution Vectors

Microsoft researchers have disclosed a critical exploit chain dubbed AutoJack that allows a single malicious web page to hijack an AI browsing agent and execute arbitrary code on the host machine. The attack works by directing an AI agent to load an attacker-controlled page, whose JavaScript can then reach privileged local services running on the same machine and spawn unauthorized processes.

The exploit abuses trust in localhost connections, missing authentication mechanisms, and unsafe parameter handling within AutoGen Studio's MCP WebSocket interface. No user interaction beyond submitting a URL is required. The research highlights a broader pattern of AI-agent localhost-hijack attacks that has been documented throughout 2026, with similar vulnerabilities previously identified in Microsoft's own Semantic Kernel framework.

Security practitioners running AI browsing agents locally are advised to audit exposed local interfaces and enforce least-privilege network policies to mitigate this class of attack.

Source: Microsoft Security Blog | The Hacker News

Avalon Malware Framework Embeds CrownX Ransomware, Targets Backup Systems

Cybersecurity researchers have identified a previously undocumented modular malware framework codenamed Avalon, distributed through sophisticated multi-stage phishing campaigns designed to bypass traditional security controls. Avalon combines credential collection, lateral movement, remote access capabilities, recovery system disruption, and ransomware deployment into a single coordinated toolkit.

The framework features an internally labeled ransomware component called CrownX that specifically targets backup and recovery systems, making data restoration significantly more difficult for victims. This approach of neutralizing recovery infrastructure before deploying ransomware represents an escalation in ransomware operational sophistication, as organizations lose their primary defense against encryption attacks.

Avalon is particularly concerning for small and medium-sized businesses that may lack the layered security controls needed to detect and block each stage of the multi-phase attack chain.

Source: The Hacker News | CyberPress

Anthropic Explores Custom AI Chip Manufacturing With Samsung

Anthropic has entered preliminary discussions with Samsung Electronics to manufacture a custom AI accelerator chip, reportedly leveraging Samsung's advanced 2-nanometer SF2P process and in-house HBM memory production capabilities. The news, reported by The Information on July 2, comes about a week after OpenAI announced its own custom AI chip partnership with Broadcom, signaling a broader industry shift toward in-house silicon design among leading AI companies.

Anthropic has recently hired specialized silicon engineers and is still defining chip specifications, power requirements, and server cluster integration details. No final design, target workload, or performance specifications have been confirmed. The move would reduce dependence on third-party GPU suppliers and potentially lower inference costs at scale.

This development adds to the growing trend of AI companies pursuing vertical integration into hardware, joining Google, Amazon, Meta, and Microsoft in building custom silicon tailored to their specific workloads.

Source: TechCrunch | Android Headlines

Tesla Launches Fully Unsupervised Robotaxi Service in Miami

Tesla rolled out its Robotaxi service in Miami on July 3, 2026, marking the first fully unsupervised, driverless ride-hailing deployment outside Texas. Unlike previous launches in other cities that included safety monitors, Miami's launch features vehicles with no human inside at all — a significant milestone in autonomous vehicle commercialization.

The service covers approximately 10 to 14 square miles in western Miami-Dade County, accessible through Tesla's app-based booking system. The expansion brings the total number of Robotaxi territories to five, operating on three different trust settings across the United States. Tesla aims to expand the service to a dozen US states by the end of 2026.

The launch has drawn attention for operating in a market with heavy rainfall, posing a notable challenge for Tesla's camera-only Full Self-Driving system. The service comes amid ongoing reliability and safety questions that regulators and the public are closely monitoring.

Source: Good Car Bad Car | Hoodline

Black Duck Report: Open Source Vulnerabilities Double as AI Accelerates Code Creation

The 2026 Open Source Security Risk Analysis (OSSRA) report from Black Duck reveals a dramatic increase in open source vulnerabilities, with the mean number of vulnerabilities per codebase jumping 107% year-over-year to an average of 581. Open source component counts also increased 30%, and the number of files per codebase grew 74%, significantly expanding the attack surface for development teams.

The report attributes much of this surge to AI-driven code generation tools that accelerate software development but also introduce vulnerabilities at scale. According to the findings, 87% of codebases are now at risk from open source vulnerabilities, and 65% have been hit by attacks exploiting these dependencies.

The data underscores the growing importance of automated dependency monitoring, software bill of materials (SBOM) management, and continuous vulnerability scanning as AI-generated code becomes more prevalent in production environments.

Source: Black Duck | PR Newswire

Google Cloud Holds Inaugural Cloud Summit in Africa, Announces Major Infrastructure Investments

Google Cloud hosted its first-ever Cloud Summit in Africa at the Sandton Convention Centre in Johannesburg on July 1, 2026, bringing together more than 3,000 business leaders, developers, public sector officials, and partners. South African President Cyril Ramaphosa opened the event, which centered on the theme of digital transformation across the continent.

During the summit, Google announced five major AI initiatives for Africa spanning research, digital skills training, startup funding, and cloud infrastructure investment. The announcements include a new applied AI lab, expanded connectivity to the continent, and long-term infrastructure commitments aimed at supporting the growth of AI and cloud services across African markets.

The summit signals Google's strategic push to expand its cloud footprint in emerging markets, competing with other hyperscalers for influence in one of the world's fastest-growing technology regions.

Source: Google Cloud Press Corner | TechArena