Bloomberg reports that Fireworks AI, an Nvidia-backed startup that runs AI models for other companies, told customers on Monday that attackers had accessed their data. According to Bloomberg, this is the company's second known hack in a few weeks. The excerpt doesn't say what data was exposed or how many customers were affected.
Why it matters. Companies that send prompts, data, or models to hosted inference providers rely on those vendors to keep it secure. Two breaches close together should push Fireworks customers to review their exposure and how much they depend on the vendor.
For Engineering and security teams that use hosted AI inference platforms, especially Fireworks AI customers. Check for a statement from Fireworks AI on what data was accessed, how many customers were affected, how the attackers got in, and whether the two incidents are connected.
AWS says Amazon Bedrock Managed Knowledge Base now handles single-shot RAG and retrieval-only requests through AgenticRetrieveStream, and agentic retrieval is still available. AWS says a parameter called maxAgentIteration controls how much retrieval effort each request gets. Its walkthrough uses a made-up internal banking policy assistant as the example.
Why it matters. Teams building RAG on Bedrock can match retrieval depth to each query. Simple lookups can use a cheap single pass, and multi-step agentic loops can be saved for harder questions, which can help them balance cost and latency against answer quality.
For AWS developers and ML engineers building RAG and agent applications on Amazon Bedrock. Check the AWS docs for which regions support AgenticRetrieveStream, how it's priced, and the default and maximum values for maxAgentIteration.
AWS says NVIDIA's Nemotron 3.5 Super VL, a 120B-parameter vision-language model, can now be deployed from SageMaker JumpStart. Teams can choose an NVFP4 or BF16 variant and run inference on text, images, and video. AWS is positioning it for document understanding, computer-use agents, and other multimodal agent work.
Why it matters. AWS teams get a managed way to try a large open NVIDIA multimodal model for agent and document workflows without building their own serving setup.
For ML engineers and AWS-based teams building multimodal or agentic applications. Check region availability, the instance types each variant needs, pricing, and license terms. Also look for independent benchmarks comparing the NVFP4 and BF16 variants.
A cloud agent can scaffold an app, add Postgres and Cursor sign-in, and deploy it after security checks. You need a paid plan with cloud agents. Nothing goes live until you explicitly publish.
Why it matters. A cloud agent can scaffold and deploy an app, and nothing goes live until you explicitly publish.
For paid Cursor users who want a hosted app. Recheck if publish stops being a required step or the paid-plan requirement changes.
Corrections: if a fact is wrong, email hello@loadedtokens.com.
What happened
Published news
No published issue in this topic yet. We do not backfill old news to make the shelf look busy. The explainers and profiles below are the context we do have.
