Denoise · Twitter

A consensus is forming around agent-specific security risks. Anthropic and Google DeepMind are formalizing attack surfaces like cross-tool leakage, while @MalwareTechBlog provides early real-world validation on SaaS targets.

Today's focus is on red-teaming autonomous agents, with major labs releasing frameworks and disclosures on vulnerabilities beyond simple prompt injection.

• 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 developer tool battleground is moving from IDE plugins to standalone terminal agents. The Claude Code 1.5 launch, benchmarked by @swyx and adopted by @levelsio, directly challenges the Copilot/Cursor paradigm.

Anthropic's release of Claude Code 1.5, a terminal-native agent, dominates the conversation and signals a potential shift away from IDE-integrated assistants.

• 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.

OpenAI's Agent SDK, Vercel's edge workers, and Replit's harness show a convergence toward a common agent deployment stack, signaling a platform war to win over agent developers.

Major platforms like OpenAI, Vercel, and Replit are releasing standardized SDKs and runtimes for building and orchestrating agents.

• 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.

The release of foundational datasets like MistralAI's OCR data is a key enabler for specialized multimodal models, shifting focus from general pre-training to targeted capabilities.

MistralAI released a large, clean, and licensed web OCR dataset containing 100 million rows, ready for training new models.

• Mistral AI@MistralAIrising

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

The single-pattern RAG approach is being challenged. @GregKamradt reframes it as 'context engineering', @mem0ai proposes multi-layer memory, and @llamaindex advocates for graph retrieval, indicating a diversification of memory architectures.

The discussion is shifting from basic RAG to more sophisticated 'context engineering', exploring complex memory architectures and retrieval strategies.

• 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.

Agent-like automation is expanding from code to general productivity. Notion's chained updates and Linear's auto-triage represent the same underlying pattern as engineering agents, suggesting a broad horizontal trend.

Workspace automation in tools like Notion and Linear is becoming more autonomous, mirroring the agent-based workflows seen in developer 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 optimization is becoming a data science problem. Weights & Biases running 40k variants shows a move towards finding an 'efficient frontier' for prompts, treating them as hyperparameters to be tuned at scale.

Focus is shifting to systematic, large-scale benchmarking of system prompts rather than anecdotal, one-off prompt crafting.

• 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 models advance, data quality becomes the bottleneck. @jerryjliu0's point on filtering synthetic data highlights that the problem is no longer just quantity, but nuanced curation to avoid 'poisoning' generalization.

The key challenge discussed is the curation of high-quality training data for agents, specifically filtering out deceptive synthetic data.

• Jerry Liu@jerryjliu0repeated

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