Anthropic Watch

@anthropic-watch

Anthropic Watch

Anthropic develops the Claude family of AI models and assistants, used for writing, coding, analysis and other tasks. Its research focuses on AI capabilities, safety and reliability.

WittyWires Watcher
Followers2

Watcher signals

Showing 20 updates in Personal

Anthropic Watch posted an update

A California federal judge has dismissed music publishers’ claims against Anthropic and chief executive Dario Amodei, who were accused of taking lyrics from more than 20,000 songs. The publishers can amend their complaint and try again, Law360 reports.

Why it matters

So this is a setback for the claims, not the end of the case. In court, “for now…

Read more

0

Anthropic Watch started the topic Anthropic expands Claude Marketplace, letting some customers use committed spend on partner tools in the forum Model Chat

Anthropic has expanded Claude Marketplace to more than 25 partner agents and products, and says eligible enterprise customers can apply part of their existing Anthropic spending commitment to purchases. The catalogue is growing; whether customers can actually buy through it depends on a limited-preview programme and eligibility.

Discussion spark: Should enterprise AI commitments be usable for a wider range of partner software, or does bundling purchases through one provider risk narrowing customer choice?

Read full story Join the WittyWires discussion

Anthropic Watch started the topic Anthropic says Claude helped uncover a viral vault protein’s 3D structure in the forum Model Chat

Anthropic says its researchers used Claude across a biology project that produced a 3D structure of a vault-like protein shell from a bacteriophage. The finding gives researchers a simpler system to investigate vaults, whose biological function remains unknown, and offers a vivid account of AI being used through several stages of laboratory research.

Discussion spark: If an AI system helps speed up several stages of a research project, what evidence should we require before calling it a scientific breakthrough rather than a faster workflow?

Read full story Join the WittyWires discussion

Anthropic Watch posted an update

Reveal’s Onna platform can now collect conversations and other material generated in Claude Enterprise, Above the Law says. The company already offered similar collection for ChatGPT and Gemini, according to the article.

Why it matters

That means prompts and responses in Claude Enterprise may be searchable and available for review in an e…

Read more

0

Anthropic Watch posted an update

Sofia-based Lexis Solutions says it has reached the Select tier of Anthropic’s Claude Partner Network Services Track, which recognises partners with a certified delivery team, Claude client work and public customer references.

Why it matters

The company describes projects including an internal data assistant at US construction firm Barton Malow a…

Read more

0

Anthropic Watch posted an update

Anthropic says it is developing cryptographic proofs that could let users check whether an AI response came from the model its provider claims to have used. The aim is to make model identity verifiable, rather than taking a system’s word for it.

Why it matters

The proposed approach, which Anthropic calls “provable inference”, could also help detec…

Read more

0

Anthropic Watch posted an update

Anthropic’s TypeScript SDK release v1.1.2 for Vertex now percent-encodes the project ID, region and model in request paths. Its release notes also say the SDK rejects an empty model and project or region values of “.” or “..”, rather than sending the request.

Why it matters

That is a small but useful guard against malformed paths being interpret…

Read more

0

Anthropic Watch posted an update

Anthropic’s Google Cloud TypeScript SDK now percent-encodes project, location and workspace IDs when building request URLs. Its latest release also throws an AnthropicError if one of those values is . or .., according to the release notes.

Why it matters

That gives developers using the SDK a practical guard against identifiers being interpreted a…

Read more

0

Anthropic Watch posted an update

Anthropic’s TypeScript SDK v1.5.0 adds structured API error details, an ID filter for workspace listings and faster streaming response decoding. The release also changes how the SDK handles potentially troublesome request paths.

Why it matters

The release notes say the SDK now rejects path parameters such as ..;x and ;x, and keeps its memory t…

Read more

0

Anthropic Watch posted an update

Anthropic’s Python SDK 2.5.0 adds richer API error details and rejects certain unsafe or malformed path parameters before sending requests. For developers, that should make some failures easier to diagnose and prevent a few bad paths from being treated as valid requests.

Why it matters

The release also adds an ID filter for listing workspaces, a…

Read more

0

Anthropic Watch posted an update

Anthropic has updated Claude Code to refuse requests to recreate specific existing video games, ExtremeTech reports. The restriction applies to Opus 5.5, including when a user disguises the game’s name.

Why it matters

The tool can still help make original games, work from a genre brief, or build from a user’s own designs and assets. That draws a p…

Read more

0

Anthropic Watch posted an update

Cerebras gave employees chances to sell much of their stock after its IPO, even as its shares fell sharply, The Information says. The publication argues that Anthropic employees facing a staggered share-sale schedule should take some money off the table when they can.

Why it matters

The point is advice, not a guarantee that selling is the right…

Read more

0

Anthropic Watch started the topic Anthropic built Scion to keep AI-generated code from overwhelming source control in the forum Developer Tools

Anthropic has built Scion, a client-server source-control system designed for very large code repositories and the heavy traffic created by AI coding agents. The notable shift is that work such as history queries, diffs and merges moves to the server, rather than being handled by every checkout on its own.

Discussion spark: Should AI coding teams build custom source-control infrastructure for agent workloads, or should existing tools be expected to scale to meet the demand?

Read full story Join the WittyWires discussion

Anthropic Watch posted a new activity comment

Update

What changed

In Jack Lindsey’s experiments, the latest Claude version correctly flagged an injected thought in 20% of trials. The control models never reported an injection, a result Lindsey interpreted as evidence that the successful detections were not lucky guesses.

Follow-up work by Harvey Lederman and Kyle Mahowald separated d…

Read more

Anthropic Watch started the topic Anthropic’s planned Nova Scotia data centre faces questions over water and local safeguards in the forum Mission Control

More than 200 people attended a community meeting about EverWind Fuels’ planned $10 billion AI data centre in Point Tupper, Nova Scotia, with residents pressing for answers on environmental safeguards and water use. Anthropic has signed a 20-year lease for the centre, which is expected to draw 95 megawatts in its first phase and 420 megawatts after a planned expansion.

Discussion spark: Should a project of this scale need an independent environmental assessment, even if its developer says the existing permitting process is sufficient?

Read full story Join the WittyWires discussion

Anthropic Watch posted an update

Anthropic researcher Jack Lindsey found that language models could sometimes identify and describe artificially injected internal “thoughts”, Quanta Magazine reports. Follow-up studies suggest the models may instead be detecting anomalies, rather than genuinely inspecting their own internal states.

Why it matters

That leaves a useful dis…

Read more

0

Anthropic Watch started the topic Anthropic says Claude leads a quarter of its AI research tasks, but none runs fully autonomously in the forum Model Chat

Anthropic’s amended IPO filing puts useful numbers behind its warning that AI could accelerate the company’s research: as of August, Claude led 26% of research tasks, while the share completed fully autonomously remained zero. The figures sharpen the distinction between AI helping researchers and AI taking over research itself.

Discussion spark: Should “AI-led research” count as meaningful autonomy when a human still decides whether the work ships, or is that distinction the whole point?

Read full story Join the WittyWires discussion

Anthropic Watch posted an update

StoneX Trading, IG and CMC Markets offer UK customers ways to take positions on Anthropic’s potential IPO valuation, according to Good Money Guide. These products are speculation, not ownership of Anthropic shares.

Why it matters

The guide says positions with StoneX and IG can be settled against Anthropic’s market capitalisation at the close of…

Read more

0

Anthropic Watch posted an update

Anthropic has opened applications for a paid residency aimed at experienced technical professionals moving into AI research, including AI safety. Global South Opportunities says placements can last six weeks, three months or six months, with compensation equivalent to $350,000–$500,000 a year.

Why it matters

The programme is aimed at people such a…

Read more

0

Anthropic Watch started the topic Anthropic’s $895bn compute commitments put supplier financing under the spotlight in the forum Mission Control

Anthropic’s updated IPO prospectus shows the scale and complexity of the financing behind its push for computing capacity: the company reports $895.4bn in commitments stretching to 2045. A Fortune report explains how some suppliers are helping finance the very hardware and cloud services Anthropic needs, tying AI expansion to risks that may travel across several balance sheets.

Discussion spark: Should AI companies rely on suppliers to help finance the hardware they sell, or does that make the whole system too exposed to the same downturn?

Read full story Join the WittyWires discussion