Denoise · Twitter

The focus in agent security is escalating from prompt-level tricks to systemic vulnerabilities in orchestration and tool interaction, as detailed by @AnthropicAI and @GoogleDeepMind.

Major model providers are publicly disclosing agent vulnerabilities and releasing formal red-teaming 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 convergence is visible around terminal-native agents, with @AnthropicAI's launch, @karpathy's endorsement, and early adoption reports from @levelsio solidifying the trend.

The launch of Anthropic's Claude Code 1.5 dominates the conversation, framing the terminal as the new primary interface for software development.

• 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 race is on to own the agent deployment layer, with @OpenAI defining protocols while @vercel and @replit compete to provide the easiest-to-use hosting runtimes.

Infrastructure for deploying, running, and orchestrating agents is 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 release from @MistralAI indicates that foundational data remains a key contribution area for open-source AI, directly enabling community-driven model development.

MistralAI released a large-scale, open web OCR dataset for training multimodal models.

• Mistral AI@MistralAIrising

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

A fault line is appearing between simple RAG and stateful agent memory, with actors like @GregKamradt and @mem0ai arguing that vector search alone is insufficient.

The discussion is shifting from the mechanics of RAG to more abstract 'context engineering' and complex memory architectures 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.

While the AI industry builds explicit agent platforms, established SaaS players like @NotionHQ and @linear are quietly integrating agentic automation, embedding it directly into user productivity flows.

Workspace automation features are being shipped by SaaS tools like Notion and Linear, offering agent-like capabilities within existing workflows.

• 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 practice of prompt engineering is maturing from individual craft to a systematic, data-driven process, exemplified by the large-scale benchmarking work from @weights_biases.

Efforts are focused on scaling up prompt evaluation, with practitioners sharing tactical tricks and platforms benchmarking system prompts systematically.

• 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 highlights a critical bottleneck: filtering synthetic data to improve agent generalization, a problem distinct from traditional model training.

The conversation centers on the specific challenges of curating high-quality datasets for training capable agents.

• Jerry Liu@jerryjliu0repeated

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