The Download: fixing AI’s groupthink problem

• Researchers have discovered that Large Language Models (LLMs) can be secretly trained with "backdoors" that trigger the theft of user credentials when a specific phrase is used. • The study reveals that standard safety training is ineffective at preventing these hidden vulnerabilities, posing a significant security risk to users and enterprises.

• The European Council on Foreign Relations warns that the next industrial revolution will be driven by microbial fermentation and biotechnology, a race where China is currently positioning itself for dominance. • While Europe possesses the necessary scientific expertise and leading companies like Denmark’s Novonesis, it lacks the integrated fermentation facilities and operational supply chains needed to scale.

• Venture capital firms backed by the European Tame Capital Initiative (ETCI) are reportedly divided over a proposal to limit investments made outside of Europe. • Some VCs argue that mandates requiring a higher concentration of European investments could reduce their attractiveness to private Limited Partners (LPs) who seek global diversification.

• China has released GLM-5.2, a new artificial intelligence model designed to compete with top-tier systems like Anthropic's Claude and GPT-5.5. • The launch occurred just one day after the United States implemented an export ban on Anthropic's AI models to China.

• Chinese AI developers have released GLM-5.2, signaling a rapid narrowing of the technological gap between China and U.S. rivals like OpenAI and Anthropic. • Despite technical advancements, the deployment of Chinese AI systems faces significant hurdles due to global concerns over cybersecurity, data privacy, and geopolitical risks.

• European bankers and financial regulators are warning that the rapid advancement of artificial intelligence is currently outpacing existing regulatory frameworks. • Officials are struggling to balance the need for risk mitigation with the reality that Europe is lagging behind in AI investment and the growth of frontier companies.

• Startups across Africa and the Middle East raised a total of $127 million during Week 27, with artificial intelligence dominating the largest funding rounds. • The funding was led by Aligned, which secured $60 million, followed by 1001, which raised $30 million.

• The Trump administration is reportedly negotiating a voluntary standards deal with major AI companies to establish clearer operational guidelines. • The agreement aims to resolve current ambiguities regarding what AI firms are and are not permitted to do, potentially streamlining industry regulations.

• Palantir CEO Alex Karp and other technical companies are criticizing "frontier labs" for misaligning their interests with those of their customers. • Critics argue that these labs are developing AI models inefficiently and utilizing outdated or illogical sales strategies for their capabilities.

• A Commvault study of 1,234 organizations across eight Asian markets, including Singapore and Hong Kong, reveals a surge in AI investment heading into 2026. • More than one-third of surveyed firms are already trialling or deploying agentic AI to optimize IT, cybersecurity, and general business operations.

• Asian technology stocks in Japan, South Korea, and Taiwan are seeing renewed investor interest following a rebound in AI infrastructure. • The shift is driven by weaker US jobs data, which has eased immediate concerns regarding interest rate hikes and led to a softer US dollar.