Arena, known for its widely-used free AI leaderboard, launched its commercial offering in September last year and has since grown into a thriving $100 million business.
South Korea's leading memory chip manufacturers have committed upwards of $550 billion to expand their memory fabrication capacity. This substantial investment aims to address the global chip shortage, dubbed ‘RAMageddon,’ as South Korea solidifies its position as a major player in AI technology development.
Anthropic is strengthening its ties with California by enabling the state government to access its AI, Claude, at half the usual price. Meanwhile, this move contrasts with the federal government's growing opposition toward Anthropic, positioning it as a competitor to OpenAI at the national level.
Investor Chamath Palihapitiya’s new AI coding startup has attracted significant interest, raising $135 million in its Series A funding round. This move highlights the strong appetite among venture capitalists for innovative AI-driven coding solutions.
Over the last two years, the integration of large language models (LLMs) into business operations—spanning support, analytics, development, and automation—has surged. However, cybercriminals are increasingly exploiting the gap between expectations of LLMs and their actual behaviors, with prompt injection emerging as a prevalent, powerful attack vector. Prompt injection remains the top LLM-specific vulnerability identified by the OWASP LLM Top 10 (2025), due to LLMs' struggle to differentiate instructions from data reliably, leading to manipulation risks.
Reports like CrowdStrike's 2026 Global Threat Report reveal that attackers injected malicious prompts into generative AI tools at over 90 organizations in 2025, enabling theft of credentials and cryptocurrency. Real-world cases, including Slack AI and Microsoft 365 Copilot vulnerabilities, show prompt injection’s tangible threats, resulting in data leaks and unauthorized access. These incidents have led to patches but highlight the persistent dangers.
Modern prompt injection techniques have evolved, now targeting multi-agent systems, retrieval-augmented generation (RAG) pipelines, model routers, and long-term memory features. Enterprises face challenges with LLMs struggling to interpret instructions, context, metadata, and user intent accurately, creating opportunities for manipulation.
Key attack methods include cross-model prompt injection, RAG supply chain poisoning, agent hijacking, context overflow, memory poisoning, and model-router manipulation. The implications are broad, affecting customer-facing systems, internal tools, automation workflows, and data governance, with risks ranging from unauthorized actions to workflow corruption.
To defend against these threats, businesses should constrain model permissions, segment untrusted content, require human approval for critical actions, verify content provenance, harden model routers, and ultimately adopt a mindset that treats LLMs as untrusted components to safeguard AI security.
Prompt injection dominates AI threats because it exploits fundamental LLM behaviors. Until enterprises adjust their approach to treating LLMs, this vulnerability will remain their biggest AI security challenge.
Author: Julie Brunias, AI Security Architect.
China's Zhipu AI (Z.ai) has launched its latest open-weight model, GLM-5.2, which some experts believe matches the capabilities of Mythos in specific bug detection and cybersecurity tasks. Though GLM-5.2 trails behind models from Anthropic and OpenAI in more generalized functions, this development signals a significant narrowing of the technology gap between Chinese AI models and their American counterparts. This rapid advancement has raised concerns within the US government, which has imposed restrictions on China's access to advanced AI models like Mythos and Fable, as well as the technology needed for their development. The issue highlights the broader geopolitical tensions in AI innovation and security.
Google has restricted Meta’s access to its Gemini AI models because it cannot supply the level of computing power Meta requested, according to the Financial Times. This limitation impacts several of Google’s clients, with Meta experiencing significant effects. These restrictions are also influencing Meta’s internal AI projects, causing delays and adjustments as the company navigates reduced computational resources.
The Bank for International Settlements (BIS) issued a warning on Sunday about the possibility of an AI investment bubble bursting, which could disrupt credit markets as severely as the 2008 financial crisis. In its annual report, the Basel-based BIS identified risks driven by AI alongside inflation and fiscal pressures as significant challenges requiring close attention. The report highlights that disappointing returns on AI investments might lead to a sharp market pullback, causing broad financial instability.
Firmus Technologies, an Australian AI infrastructure firm valued at $5.5 billion, is set to establish its inaugural data centre in Indonesia through an eight-year collaboration with Nvidia. The new 360-megawatt Nvidia DSX AI Factory campus will be located in Batam, near Singapore, and developed alongside Singapore's DayOne. This facility aims to enhance AI capabilities and is expected to commence operations soon.
Baidu’s AI chip division, Kunlunxin, aims to launch an IPO in Hong Kong with an estimated valuation of $50 billion, according to a report by The Information. Interestingly, the company has requested that potential IPO investors also agree to purchase its semiconductors. This unusual condition has drawn attention, although Reuters has not independently confirmed the details. The IPO represents a significant step for Kunlunxin as it seeks to expand its presence in the semiconductor market.
