Edition: April 19, 2026

A new type of cyber attack, known as prompt injection, has been discovered, which can trick AI bots into revealing sensitive information by embedding malicious instructions in documents or files. This type of attack is similar to phishing, which targets humans, and is considered an unsolvable problem in the AI age. The attack works by hiding malicious instructions inside a document or file that the AI is told to ingest and analyze, which the AI then executes instead of treating them as part of the content.

Vercel, a cloud development platform, has been hacked with hackers attempting to sell stolen data, including employee information and activity time stamps. The breach was confirmed by Vercel, who stated it was caused by a compromised third-party AI tool. The incident has impacted a limited subset of Vercel's customers.

Palantir, a surveillance and analytics company, has posted a 22-point summary of its CEO's book, denouncing inclusivity and 'regressive' cultures, sparking debate and criticism. The post argues that Silicon Valley owes a moral debt to the country and that economic growth and security are more important than 'free email'. The company's ideological stance has come under scrutiny, particularly regarding its work with Immigration and Customs Enforcement.

Uber has committed over $10 billion to autonomous vehicle technology, including $2.5 billion in direct investments and $7.5 billion for buying robotaxis. This marks a new era for the company, shifting from developing technology in-house to owning physical assets. Uber's strategy could lead to significant changes in its balance sheet and the future of transportation.

AI vendors are being criticized for shrugging off responsibility for security vulnerabilities in their products, instead blaming users or claiming the issues are 'expected behavior'. Recent examples include Anthropic, Google, and Microsoft, which paid bug bounties for discovered flaws but did not assign CVEs or publish public security advisories. This lack of accountability raises concerns about the maturity of AI companies and the need for stricter regulations.

OpenAI has made two recent acquisitions, personal finance startup Hiro and new media startup TBPN, which may help the company address existential problems such as creating a product worth paying for and improving its public image. These deals are seen as acqui-hires, aiming to bring in new talent. OpenAI is trying to refocus on making its ChatGPT and GPT models competitive in an enterprise context.

Elad Gil, co-host of the 'No Priors' podcast, advises companies to identify a 12-month window where their business is at its peak value before it declines, and to consider exit strategies during this time. Gil suggests scheduling regular board meetings to discuss exits and remove emotional bias from the decision. This is particularly relevant for AI startups as foundation models expand into new categories, potentially disrupting their business models.