Denoise · Twitter

The red-teaming focus is shifting from simple prompt injection to complex, stateful attacks on agent frameworks, as demonstrated by Anthropic's disclosure and DeepMind's new framework.

Major labs are publicly disclosing and building frameworks for agent security, focusing on vulnerabilities in orchestration and multi-tool systems.

• 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 primary competition in AI coding is now between integrated terminal agents like Claude Code and IDE-based tools like Copilot, a shift articulated by @karpathy and validated by @levelsio.

Anthropic's Claude Code 1.5 release intensified the move toward terminal-native coding agents, sparking immediate comparisons and adoption.

• 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.

A convergence is happening around the 'agent worker' as a core compute primitive. OpenAI is standardizing protocols, while Vercel and Replit provide competing serverless deployment targets for them.

Major platforms like OpenAI, Vercel, and Replit are racing to provide the essential infrastructure for deploying and managing autonomous 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.

Rather than releasing a model, Mistral AI is providing a foundational dataset, a strategic move to commoditize a key component for building multimodal AI and foster ecosystem development.

Mistral AI released a large-scale, cleaned web OCR dataset to the public, enabling broader research and training for multimodal models.

• Mistral AI@MistralAIrising

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

Experts like @GregKamradt and startups like @mem0ai are pushing beyond vector search, exploring caching, memory hierarchies, and knowledge graphs (@llamaindex) as necessary components for agent memory.

The discourse on information retrieval for LLMs is shifting from simple RAG to more complex 'context engineering' and structured memory systems.

• 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 pattern of agent-driven automation in coding is mirrored in productivity tools, where both Notion and Linear are building agent-like features to reduce manual user input.

Workspace automation features are becoming standard, with Notion and Linear releasing tools that automatically manage data and triage tasks.

• 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.

The work by @weights_biases, testing 40k prompt variants, shows a maturing, data-driven approach to prompt optimization that contrasts with individual heuristic sharing like @dotey's thread.

The focus in prompt engineering is moving from anecdotal tricks to systematic, large-scale benchmarking to find optimal instructions.

• 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 training becomes more common, nuanced data curation techniques, like those shared by @jerryjliu0, become critical for avoiding performance degradation from poisoned synthetic datasets.

The challenge of curating high-quality training data for agents involves filtering out synthetic data that harms model generalization.

• Jerry Liu@jerryjliu0repeated

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