Brussels tech lobbying heats up as DHS bans DeepSeek and an AI coding tool gets breached
Microsoft and Google step up pressure on EU tech rules, US federal agencies are barred from using DeepSeek, and a popular open-source coding agent suffers a breach tied to an API key theft scheme.
Every line links to a primary source in the full briefing.
01Microsoft and Google press Brussels to soften AI Act and DMA rules
Microsoft and Google are lobbying EU institutions to ease requirements under the AI Act and the Digital Markets Act, according to reporting corroborated by five outlets including the European Commission itself. The push targets compliance burdens the companies argue slow down AI deployment in Europe.
Rules that get watered down now under lobbying pressure could shift again before most small firms even finish adapting to the current version, so it is worth watching rather than rushing to comply with rules that may not stick.
Source: El Ecosistema Startup
02US bars DeepSeek from federal civilian agency systems
The Department of Homeland Security has banned the Chinese AI model DeepSeek from use across federal civilian agencies, citing security concerns, according to five outlets. The move extends a pattern of US government restrictions on Chinese-made AI tools.
Companies that work as federal contractors or in regulated supply chains should check whether similar restrictions could extend to their own vendor lists.
Source: BankInfoSecurity
03Survey: most AI professionals see the EU's AI law as an edge, not a burden
A survey by Shakers found that 63% of AI professionals view the EU's AI Act as giving European companies a competitive advantage rather than holding them back. The finding cuts against the common narrative that the law is purely a cost center for business.
Owners weighing whether to invest in compliance now have a data point suggesting early movers on governance could differentiate themselves rather than simply absorb a cost.
Source: El Ecosistema Startup
04Open-source coding agent OpenCode breached, API keys stolen
OpenCode, an open-source AI coding agent that competes with Claude Code and Cursor, said attackers broke into an older part of its platform and stole API keys from about 1.4% of users of its online console. The company suspects the keys were used to run large-scale AI output generation, though some observers say that explanation is unproven and the incident may simply be a straightforward breach.
Any business that lets employees store API keys in third-party developer tools should rotate those keys and check vendor breach notices, since a stolen key can run up large charges on your account before anyone notices.
Source: TechCrunch
05Amazon adds structured outputs for Claude models on Bedrock
Amazon Web Services rolled out structured output support for Anthropic's Claude models running on its Bedrock cloud platform. The feature lets developers get more reliably formatted responses from the model for use in applications.
Teams building internal tools on top of Claude through AWS get an easier path to consistent, machine-readable output without extra engineering work.
Source: Amazon Web Services (AWS)
06New York mayor urges residents to push back on unchecked AI rollout
New York Mayor Zohran Mamdani said the federal government has failed to regulate AI and urged city residents to demand protections rather than accept AI expansion as inevitable. He pointed to his earlier ban on AI in elementary and middle schools and a policy requiring disclosure when real estate listings use AI-generated images.
Businesses operating in cities filling the regulatory gap left by federal inaction should expect a patchwork of local AI disclosure and use rules rather than one national standard.
Source: TechCrunch
Review which API keys your team has stored in third-party AI developer tools
The OpenCode breach shows how a single leaked API key can run up unauthorized charges or expose your AI account to misuse. Take an hour this week to list every tool where staff store API keys for AI services, rotate any that are old or shared, and set spending alerts on your AI provider accounts.
Sources
6 links- 01Microsoft y Google presionan el AI Act y el DMA en la UEEl Ecosistema Startup
- 02DHS Bans DeepSeek From Federal Civilian AgenciesBankInfoSecurity
- 03Shakers: 63% de profesionales de IA ve ventaja en la ley UEEl Ecosistema Startup
- 04Structured outputs for Anthropic Claude models on Amazon BedrockAmazon Web Services (AWS)
- 05Hackers hit AI coding tool OpenCode, running up charges on users’ AI accountsTechCrunch
- 06Mamdani says feds have abdicated on AI, urges New Yorkers to push backTechCrunch