Denoise · Twitter

The security conversation is rapidly shifting from LLM prompt injection to the more complex attack surface of agent orchestration and tool interaction, with both Anthropic and Google DeepMind publishing formal frameworks.

Major labs are disclosing agent jailbreaks and releasing red-teaming frameworks, while practitioners are testing agent-based pentesting in the wild.

• 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 landscape is re-centering around terminal agents, with Anthropic's Claude Code 1.5 directly competing with the Copilot/IDE paradigm, a shift endorsed by influential figures like @karpathy.

Anthropic released Claude Code 1.5, a terminal-native agent, prompting discussion on its performance against existing tools and the broader shift away from IDE-centric workflows.

• 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 battle for the standard agent protocol is beginning, with OpenAI's new SDK and Anthropic's Model Context Protocol (via LangChain) representing competing visions for multi-agent orchestration.

OpenAI, Vercel, and Replit released new primitives for agent orchestration and deployment, focusing on SDKs, protocols, and serverless runtimes.

• 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.

Foundational model labs continue to release high-quality, open datasets like Mistral AI's OCR data, enabling the broader community to train specialized models and commoditizing a key part of the data pipeline.

Mistral AI released a large-scale, cleaned OCR dataset for training multimodal models.

• Mistral AI@MistralAIrising

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

The naive approach of 'just use RAG' is being replaced by more complex memory architectures, with @GregKamradt, @mem0ai, and LlamaIndex all proposing different layers of memory management beyond simple vector retrieval.

The focus shifts from simply increasing context window size to sophisticated 'context engineering,' including caching strategies and specialized memory layers for agents.

• 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 concept of orchestration is becoming mainstream, with workflow engines like Temporal positioning themselves for AI agents, while productivity tools like Notion and Linear implement similar, albeit simpler, automation patterns.

Workspace automation features are launching in Notion and Linear, while Temporal discusses durable workflows for orchestrating agents.

• 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 evolving into a systematic, data-driven discipline, with efforts from Weights & Biases and individual engineers demonstrating a move from anecdotal tricks to scalable, benchmark-driven optimization.

Practitioners are sharing empirical results from large-scale system prompt benchmarking and reviews of production prompts.

• 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 relies more on synthetic data, curation and filtering techniques, as highlighted by @jerryjliu0, are becoming a critical and non-obvious part of the MLOps stack to ensure model robustness.

A practitioner shared insights on the critical process of curating and filtering synthetic data to avoid poisoning agent generalization during training.

• Jerry Liu@jerryjliu0repeated

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