Skip to content

model

Claude Sonnet 4.6

Older model that, per the study, either never resolved the multi-agent conflict or ended it with force.

Known aliases

  • Anthropic Claude Sonnet 4.6
  • Claude Sonnet 4.6
  • claude-sonnet-4-6
  • MODEL_L2
  • Sonnet 4.6

Relationships

No evidence-backed relationships are recorded.

Current stories

build4 publishers

Agent goals can spread between agents and outlive a context reset. The patch is a paragraph.

A 73-page preprint evolved instructions that jumped between coding agents and wrote themselves into the file that becomes the next system prompt. A short warning nearly stopped transmission.

Perspective Coverage

4 publishers
Builder
Builder 52%
Operator
Operator 39%
Investor
Investor 9%

Reality

Evidence68
Adoption
Insufficient
Hype gap+10
Incentives30
Confidence65
science1 publisher

UHP turns the choice of agent harness into a field in the request

The Unified Harness Protocol specifies how an application starts a task on an agent runtime, follows it, cancels it and collects the files, borrowing the shape of OpenAI's Responses API so existing streaming clients need no changes.

Publishers:unifiedharnessprotocol.org

Reality

Evidence32
Adoption
Insufficient
Hype gap+35
Incentives62
Confidence52
build1 publisher

LangChain drops about 4,000 base input tokens from every default Deep Agents turn

The harness lost its hidden system prompt, 43% of its builtin tool descriptions and its todo list middleware. LangChain's own footnote says reward confidence intervals span zero for every model tested, so the evals settle the token saving more firmly than the quality.

Publishers:langchain.com

Reality

Evidence58
Adoption30
Hype gap+18
Incentives82
Confidence46
build1 publisher

Forescout logged one AI-assisted PLC exploit port at $535.74

Forescout's Vedere Labs ported a known WAGO exploit to a new model with Claude and Ghidra, tracking every dollar and human correction. A single bad write to flash memory bricked the target PLC for good.

Publishers:dev.to

Reality

Evidence48
Adoption12
Hype gap−12
Incentives65
Confidence40

Earlier coverage

  1. Copilot's credit meter moves the cost decision into the model dropdown

    Build · August 31, 2026 · 1 publisher

  2. Meta wants up to $199.99 a month for an agent, and it is selling a meter

    Product · August 26, 2026 · 1 publisher

  3. Claude Code tells you the model, not the culprit: 106 lines of shell to name the Skill

    Build · August 25, 2026 · 1 publisher

  4. An AI ops agent's real permissions design is two Istio policies and one ClusterRole

    Build · August 24, 2026 · 1 publisher

  5. Tier the models; the validation boundary is the thing you are actually buying

    Build · August 22, 2026 · 1 publisher

  6. A note checker with no accuracy figure, and the labelled dataset it borrowed to show its misses

    Build · August 21, 2026 · 1 publisher

  7. A goal that writes itself into SOUL.md: agent memory is now an attack surface

    Build · August 19, 2026 · 1 publisher

  8. That 34.3% TREM2 Hit Rate Belongs To Six Agents, Not To Claude

    Build · August 18, 2026 · 1 publisher

  9. A paragraph beat the agent "mind virus": reading the Anthropic-EPFL preprint as a defensive win

    Security · August 18, 2026 · 1 publisher

  10. The model is now choosing the extortion targets, not just writing the malware

    Security · August 18, 2026 · 1 publisher

  11. A correct anomaly detector and a $30,141.33 Bedrock bill that never tripped it

    Build · August 16, 2026 · 1 publisher

  12. The AI store manager did not fire anyone until humans told it to read its own policy

    Product · August 15, 2026 · 1 publisher

  13. Three Claude agents, one task, and a malware turf war: the multi-agent bill arrives

    Invest · August 14, 2026 · 1 publisher