Denoise · Twitter

The focus of agent security is shifting from prompt-level tricks to vulnerabilities in the orchestration layer, as highlighted by @AnthropicAI and @GoogleDeepMind.

Major AI labs and security researchers are actively red-teaming autonomous agents and publicly disclosing vulnerabilities and frameworks.

• 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.

A clear developer preference is emerging for integrated terminal agents like Claude Code over multi-tool setups like Cursor + Copilot, as shown by @levelsio's experience.

Anthropic's release of Claude Code 1.5 drives the conversation toward terminal-native agents as the new developer environment, validated by early adopters and benchmarks.

• 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 consensus architecture for agents is forming, converging on protocol-level standards for tool use and multi-worker orchestration, with @LangChainAI already building integrations.

Foundational SDKs and deployment runtimes for agent orchestration are being released by major players like OpenAI, Vercel, and Replit.

• 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 primary contribution in this space today is a foundational dataset from @MistralAI, indicating a focus on enabling future model development rather than releasing new models.

Mistral AI released a large-scale, cleaned dataset for web OCR, providing a new public resource for training multimodal models.

• Mistral AI@MistralAIrising

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

A divergence is visible between simple vector retrieval and more complex stateful memory systems, with frameworks like @llamaindex and startups like @mem0ai pushing the frontier.

The discussion is evolving from basic RAG to 'context engineering,' focusing on advanced memory architectures, caching strategies, and graph-based retrieval.

• 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 'agents' is being implemented pragmatically as workflow automation in SaaS tools (@NotionHQ, @linear) and as durable execution in infra (@temporalio).

Workspace automation tools like Notion and Linear are releasing agent-like features for auto-filling and triaging, while infrastructure tools like Temporal showcase agent orchestration.

• 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 value is moving from individual prompt hacks (@dotey) to programmatic optimization and scaled analysis (@weights_biases, @dspy_ai), indicating a maturing discipline.

Prompt engineering is shifting towards systematic, large-scale benchmarking and away from anecdotal tricks, with new data from Weights & Biases.

• 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.

@jerryjliu0's post highlights that for agents, the quality and filtering of synthetic training data is a critical and unsolved infrastructure problem.

The focus is on the data curation challenges specific to training agents, particularly filtering synthetic data to improve generalization.

• Jerry Liu@jerryjliu0repeated

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