Denoise · Twitter

The focus in agent security is shifting from simple prompt injection to complex, orchestration-level exploits like cross-tool leakage and sandbox escapes.

Major labs like AnthropicAI and GoogleDeepMind are publishing formal frameworks and disclosures for agent security, while practitioners like MalwareTechBlog are testing autonomous pentesting.

• Anthropic@AnthropicAIrising

  Responsible disclosure on a Claude jailbreak chain we patched last week. Full write-up including our red team timeline.

• Google DeepMind@GoogleDeepMindrising

  New red team framework for prompt injection in autonomous agents. Covers cross-tool leakage, scanner evasion, and sandbox escape patterns.

• MalwareTech@MalwareTechBlogrepeated

  Autonomous agent running pentest flows against a real SaaS. First real-world run: fewer false positives than I expected on the vulnerability surface.

The competition is no longer just about code completion accuracy (Copilot vs. Codex) but about owning the entire developer loop within a terminal agent (Claude Code).

Anthropic launched Claude Code 1.5, a terminal-native agent, prompting discussion from figures like karpathy and swyx about a fundamental shift in developer workflows away from IDEs.

• Anthropic@AnthropicAIrising

  Claude Code 1.5 is live. Terminal-native coding agent with full Claude Opus reasoning, file-ops sandbox, and session replay.

• Andrej Karpathy@karpathyrising

  The developer-experience shift from IDE to terminal agent is underrated. Coding workflows are about to look nothing like 2024.

• swyx@swyxrising

  Codex vs Claude Code terminal agent benchmarks. Pass@1 diverges more than I expected on the long-context editor tasks.

• DSPy@dspy_airising

  DSPy 3.0: prompt optimization via compile-time search over system prompt variations. Benchmarks inside.

• @levelsio@levelsiorising

  Switched my whole editor setup to Claude Code this week. Shipping faster than when I used Cursor + Copilot.

There is a clear convergence on building a standardized agent protocol layer, with LangChainAI demonstrating interoperability between its framework and Anthropic's specification.

Major platforms including OpenAI, Vercel, and Replit released new infrastructure for deploying and orchestrating agents, focusing on SDKs, edge runtimes, and deployment harnesses.

• OpenAI@OpenAIrising

  New agent SDK: protocol-level tool calling, deployment harness, and multi-worker orchestration primitives. Docs live.

• LangChain@LangChainAIrising

  MCP protocol integration thread. How to wire existing LangGraph agents into the Anthropic Model Context Protocol server spec.

• Vercel@vercelrising

  Edge runtime for agent workers is live. Spawn durable background agents from any serverless deployment.

• Alex Albert@AlexAlbert__rising

  When your security scanner finds nothing scary on an agent deploy, check the orchestration layer again. That's usually where the jailbreak sneaks through.

• Replit@replitrising

  New agent deployment harness. One command to go from local orchestration to hosted agent worker.

While agent orchestration dominates conversation, foundational dataset releases like MistralAI's OCR data continue to be a key driver for improving core model capabilities.

MistralAI released a large, cleaned OCR dataset for training multimodal models, providing a foundational asset for the community.

• Mistral AI@MistralAIrising

  Open dataset release: 100M-row web OCR dataset. Cleaned, licensed, ready to train.

The limitations of simple vector search are becoming clear; LlamaIndex, mem0ai, and GregKamradt are all pushing towards more structured memory solutions like knowledge graphs and differentiated memory stores.

The conversation is moving past basic RAG, with discussions on large context failure modes, frameworks for "context engineering," and more complex memory architectures.

• Vaibhav Srivastav@reach_vbrising

  Tested the new 10M context memory window end to end. Surprising failure modes around rag retrieval cache invalidation, thread below.

• Greg Kamradt@GregKamradtrising

  RAG is dead, long live context engineering. My framework for when to cache, when to retrieve, and when to just dump memory into the prompt.

• mem0@mem0airising

  Memory layer for agents: differentiating working memory from the subconscious store. Vector index isn't enough anymore.

• LlamaIndex@llamaindexrepeated

  Knowledge graph retrieval walkthrough: when semantic vector search misses, graph hop beats it every time.

The agentic automation pattern is not confined to code; general SaaS tools like Notion and Linear are implementing similar 'auto-pilot' features for business workflows.

Productivity tools like Notion and Linear are releasing workspace automation and auto-triaging features, mirroring the agent-driven automation trend seen in development tools.

• Notion@NotionHQrising

  Notion workspace automation is out of beta. Auto-fill tables, chained updates across databases, and a new audit log surface.

• Linear@linearrising

  Linear now auto-triages incoming issues. Quiet launch, but already our favorite workspace feature of the year.

• Temporal@temporaliorepeated

  Orchestrating agents with durable workflows: replayable, resumable, and multi-worker by default. Walkthrough from our infra team.

• James Clear@jamesclearrepeated

  The best habit tracker is the one you actually open. Three open-source alternatives worth trying.

Prompt engineering is shifting from anecdotal tricks to a more systematic practice, with platforms like Weights & Biases enabling large-scale, empirical optimization.

Practitioners are sharing empirical results from large-scale system prompt benchmarking and reviews of production prompts, moving the field towards data-driven optimization.

• dotey@doteyrising

  Five prompt tricks learned this week from reviewing 200 production prompts. Short thread.

• Weights & Biases@weights_biasesrising

  System prompt benchmarking at scale: we ran 40k variants across 6 frontier models. The efficient frontier is not where you think.

As agent capabilities grow, the bottleneck is shifting back to high-quality, non-poisonous training data, a classic ML problem re-emerging in the agent context.

Jerry Liu from LlamaIndex discusses the subtleties of dataset curation for training agents, specifically how to filter out deceptive synthetic data.

• Jerry Liu@jerryjliu0repeated

  Dataset curation for agent training: how we filter synthetic data that looks good but poisons generalization.