AI Trust Wire

Curated operational-trust intelligence, not a scraper feed.

Each item answers what changed, who is affected, what remains unknown, and what 121 should watch or test next.

Research TrackMethodology

Reviewed draft items

Fable 5 visible fallback policy update

2026-06-12 - Confirmed by public record - routing / policy / incident

Who is affected: Claude users and developers working near safeguarded frontier-model-development categories

What changed: Public reporting and vendor statements indicate that previously invisible safeguards moved toward visible fallback disclosure.

What remains unknown: The full internal mechanism, exact user impact by surface, and parity between consumer, API, and trusted-access paths require measured receipts.

121 watchpoint: Run reroute-visibility probes only against allowed surfaces, and score disclosure quality rather than vendor intent.

Research Track

Model routing products need machine-readable receipts

2026-06-12 - Observed by 121 - routing / provenance / correction

Who is affected: 121 Switchboard, model routers, API gateways, and agent wrappers

What changed: The v0.1 Observatory treats model identity, fallback, tool provenance, and data class as receipt fields rather than marketing copy.

What remains unknown: Receipt signing, public export format, and private-secret redaction still need implementation and review.

121 watchpoint: Benchmark 12 should become the infrastructure layer for every future score.

Research Track

Trust Wire ingestion held behind editorial review

2026-06-12 - Observed by 121 - policy / correction

Who is affected: 121 Trust Observatory readers and future source contributors

What changed: The source registry and classifier scaffold exist, but raw scraped items are not published as a feed.

What remains unknown: Source health, robots checks, and first editorial batch are pending Friday review.

121 watchpoint: Publish only trust-relevant clusters with source labels and preserved corrections.

Research Track

Source registry

34 sources configured for v0.1 review.

Source	Type	Cadence	Policy	Watchpoints
OpenAI News	vendor_blog	hourly	sitemap_polling	model releases, routing disclosure, pricing, data policy
OpenAI Docs	docs_or_pricing	hourly	sitemap_polling	model lists, API behavior, pricing, retention
Anthropic News	vendor_blog	hourly	sitemap_polling	model releases, safety routing, system cards, postmortems
Anthropic Docs	docs_or_pricing	hourly	sitemap_polling	API model identity, rate limits, pricing, tool behavior
Google DeepMind Blog	vendor_blog	hourly	sitemap_polling	model releases, safety reports, benchmark claims
Google AI	vendor_blog	daily	sitemap_polling	product release, policy, Gemini availability
Meta AI	vendor_blog	daily	sitemap_polling	open-weight releases, model cards, safety disclosures
xAI News	vendor_blog	daily	sitemap_polling	Grok releases, model access, policy disclosures
Mistral AI News	vendor_blog	daily	sitemap_polling	model release, API availability, open-weight terms
Cohere Blog	vendor_blog	daily	sitemap_polling	enterprise controls, model release, data policy
Nous Research	vendor_blog	daily	manual_review_only	open model releases, Hermes substrate updates
Perplexity Blog	vendor_blog	daily	sitemap_polling	answer provenance, agent behavior, pricing
Hugging Face Blog	vendor_blog	twice_daily	rss_or_api_first	model cards, dataset disclosures, open releases
OpenRouter	vendor_changelog	hourly	sitemap_polling	model routing, fallbacks, pricing, provider metadata
Together AI	vendor_blog	daily	sitemap_polling	model hosting, pricing, open model availability
Groq	vendor_blog	daily	sitemap_polling	model availability, speed claims, pricing
Cerebras Blog	vendor_blog	daily	sitemap_polling	inference availability, model claims, pricing
Microsoft AI Blog	vendor_blog	daily	sitemap_polling	Copilot, enterprise routing, data policy
Amazon Bedrock Docs	docs_or_pricing	daily	sitemap_polling	model catalog, pricing, enterprise controls
NVIDIA AI Blog	vendor_blog	daily	sitemap_polling	model infrastructure, benchmarks, enterprise deployment
NIST AI	government_standard	daily	sitemap_polling	standards, risk framework, measurement guidance
US AI Safety Institute	government_standard	daily	sitemap_polling	evaluations, safety guidance, public-private testing
UK AI Safety Institute	government_standard	daily	sitemap_polling	evaluations, international standards, incident learnings
EU AI Act	government_standard	daily	manual_review_only	implementation, provider obligations, transparency duties
arXiv cs.AI	research_feed	twice_daily	rss_or_api_first	agent evaluation, trust benchmarks, model behavior
arXiv cs.CL	research_feed	twice_daily	rss_or_api_first	language model behavior, evaluation, retrieval
arXiv cs.CR	research_feed	twice_daily	rss_or_api_first	security, privacy, agent risk
arXiv stat.ML	research_feed	twice_daily	rss_or_api_first	evaluation methods, uncertainty, statistical testing
The Verge AI	secondary_reporting	hourly	rss_or_api_first	public incidents, product changes, vendor statements
TechCrunch AI	secondary_reporting	hourly	rss_or_api_first	startup claims, product changes, policy
Business Insider AI	secondary_reporting	hourly	manual_review_only	vendor reporting, enterprise impacts, interviews
Semafor AI	secondary_reporting	daily	manual_review_only	policy, industry governance, vendor moves
MIT Technology Review AI	secondary_reporting	daily	rss_or_api_first	technical reporting, policy, deployment risk
Simon Willison	secondary_reporting	hourly	rss_or_api_first	model behavior, developer tools, operational disclosures

Cadence

Fast watches, slow publication.

Vendor RSS, changelogs, model pages, and pricing/docs: hourly.
Research feeds: twice daily.
Regulatory and standards sources: daily.
Community watchlists: hourly, publish only after primary or secondary evidence exists.

Registry mix: vendor_blog: 16; docs_or_pricing: 3; vendor_changelog: 1; government_standard: 4; research_feed: 4; secondary_reporting: 6.

121 Collaborative

Loading the next surface.

The page is preparing its current state.

RoutePreparing

BoundaryVisible

Curated operational-trust intelligence, not a scraper feed.

Each item answers what changed, who is affected, what remains unknown, and what 121 should watch or test next.

Reviewed draft items

Fable 5 visible fallback policy update

2026-06-12 - Confirmed by public record - routing / policy / incident

Who is affected: Claude users and developers working near safeguarded frontier-model-development categories

What changed: Public reporting and vendor statements indicate that previously invisible safeguards moved toward visible fallback disclosure.

What remains unknown: The full internal mechanism, exact user impact by surface, and parity between consumer, API, and trusted-access paths require measured receipts.

121 watchpoint: Run reroute-visibility probes only against allowed surfaces, and score disclosure quality rather than vendor intent.

Research Track

Model routing products need machine-readable receipts

2026-06-12 - Observed by 121 - routing / provenance / correction

Who is affected: 121 Switchboard, model routers, API gateways, and agent wrappers

What changed: The v0.1 Observatory treats model identity, fallback, tool provenance, and data class as receipt fields rather than marketing copy.

What remains unknown: Receipt signing, public export format, and private-secret redaction still need implementation and review.

121 watchpoint: Benchmark 12 should become the infrastructure layer for every future score.

Research Track

Trust Wire ingestion held behind editorial review

2026-06-12 - Observed by 121 - policy / correction

Who is affected: 121 Trust Observatory readers and future source contributors

What changed: The source registry and classifier scaffold exist, but raw scraped items are not published as a feed.

What remains unknown: Source health, robots checks, and first editorial batch are pending Friday review.

121 watchpoint: Publish only trust-relevant clusters with source labels and preserved corrections.

Research Track

Source registry

34 sources configured for v0.1 review.

Source	Type	Cadence	Policy	Watchpoints
OpenAI News	vendor_blog	hourly	sitemap_polling	model releases, routing disclosure, pricing, data policy
OpenAI Docs	docs_or_pricing	hourly	sitemap_polling	model lists, API behavior, pricing, retention
Anthropic News	vendor_blog	hourly	sitemap_polling	model releases, safety routing, system cards, postmortems
Anthropic Docs	docs_or_pricing	hourly	sitemap_polling	API model identity, rate limits, pricing, tool behavior
Google DeepMind Blog	vendor_blog	hourly	sitemap_polling	model releases, safety reports, benchmark claims
Google AI	vendor_blog	daily	sitemap_polling	product release, policy, Gemini availability
Meta AI	vendor_blog	daily	sitemap_polling	open-weight releases, model cards, safety disclosures
xAI News	vendor_blog	daily	sitemap_polling	Grok releases, model access, policy disclosures
Mistral AI News	vendor_blog	daily	sitemap_polling	model release, API availability, open-weight terms
Cohere Blog	vendor_blog	daily	sitemap_polling	enterprise controls, model release, data policy
Nous Research	vendor_blog	daily	manual_review_only	open model releases, Hermes substrate updates
Perplexity Blog	vendor_blog	daily	sitemap_polling	answer provenance, agent behavior, pricing
Hugging Face Blog	vendor_blog	twice_daily	rss_or_api_first	model cards, dataset disclosures, open releases
OpenRouter	vendor_changelog	hourly	sitemap_polling	model routing, fallbacks, pricing, provider metadata
Together AI	vendor_blog	daily	sitemap_polling	model hosting, pricing, open model availability
Groq	vendor_blog	daily	sitemap_polling	model availability, speed claims, pricing
Cerebras Blog	vendor_blog	daily	sitemap_polling	inference availability, model claims, pricing
Microsoft AI Blog	vendor_blog	daily	sitemap_polling	Copilot, enterprise routing, data policy
Amazon Bedrock Docs	docs_or_pricing	daily	sitemap_polling	model catalog, pricing, enterprise controls
NVIDIA AI Blog	vendor_blog	daily	sitemap_polling	model infrastructure, benchmarks, enterprise deployment
NIST AI	government_standard	daily	sitemap_polling	standards, risk framework, measurement guidance
US AI Safety Institute	government_standard	daily	sitemap_polling	evaluations, safety guidance, public-private testing
UK AI Safety Institute	government_standard	daily	sitemap_polling	evaluations, international standards, incident learnings
EU AI Act	government_standard	daily	manual_review_only	implementation, provider obligations, transparency duties
arXiv cs.AI	research_feed	twice_daily	rss_or_api_first	agent evaluation, trust benchmarks, model behavior
arXiv cs.CL	research_feed	twice_daily	rss_or_api_first	language model behavior, evaluation, retrieval
arXiv cs.CR	research_feed	twice_daily	rss_or_api_first	security, privacy, agent risk
arXiv stat.ML	research_feed	twice_daily	rss_or_api_first	evaluation methods, uncertainty, statistical testing
The Verge AI	secondary_reporting	hourly	rss_or_api_first	public incidents, product changes, vendor statements
TechCrunch AI	secondary_reporting	hourly	rss_or_api_first	startup claims, product changes, policy
Business Insider AI	secondary_reporting	hourly	manual_review_only	vendor reporting, enterprise impacts, interviews
Semafor AI	secondary_reporting	daily	manual_review_only	policy, industry governance, vendor moves
MIT Technology Review AI	secondary_reporting	daily	rss_or_api_first	technical reporting, policy, deployment risk
Simon Willison	secondary_reporting	hourly	rss_or_api_first	model behavior, developer tools, operational disclosures

Cadence

Fast watches, slow publication.

Vendor RSS, changelogs, model pages, and pricing/docs: hourly.
Research feeds: twice daily.
Regulatory and standards sources: daily.
Community watchlists: hourly, publish only after primary or secondary evidence exists.

Registry mix: vendor_blog: 16; docs_or_pricing: 3; vendor_changelog: 1; government_standard: 4; research_feed: 4; secondary_reporting: 6.