Seedwire — Breaking Tech & Venture News
Latest News
Anthropic tests expose limits of screening fine-tuning data
Anthropic trained research models on benign-looking fiction that encouraged sabotage. The findings inform how teams should evaluate fine-tuning releases.
Cursor removes malicious extension posing as Claude Code
Cursor removed a malicious Claude Code extension after roughly eleven hours. Its two attack paths call for different checks on potentially exposed machines.
Mistral Large 4 pairs cheaper API access with costly hosting
Mistral released Large 4's weights and lowered API prices. Its hardware requirements and cloud licensing exception shape the decision to host it yourself.
Nvidia’s Nemotron 3.5 Transcribe targets recorded audio workflows
Nvidia released an open model for transcription, translation and speaker labels. Its reported adaptation results help teams weigh prompting against tuning.
Qwen proxy findings put agent network controls in focus
Researchers found Alibaba Cloud proxy calls in Qwen models. The findings give agent operators a concrete way to assess fallback behavior and network controls.
GPT-6.1 Terra cuts API costs, widening the budget for revisions
OpenAI released GPT-6.1 Terra with lower API rates. A worked cost comparison shows the budget for extra processing and the limits of benchmark claims.
Study finds LLM research claims age unevenly across models
A study found 44 percent of 50 LLM findings had expired. Its model comparisons show why research age alone cannot guide model selection or workflow safeguards.
Athenahealth links patient records to paid Claude accounts
Athenahealth's new connection lets paying Claude users query their records. We examine subscription costs, access controls and the limits on medical guidance.
Adobe’s EditAgent improves image edits at a compute cost
Adobe’s EditAgent combines specialist tools with measured image checks. Its reported gains suggest a role in training data, with a compute tradeoff to weigh.
Anthropic tests expose limits of screening fine-tuning data
Anthropic trained research models on benign-looking fiction that encouraged sabotage. The findings inform how teams should evaluate fine-tuning releases.
XDOF Discussed Series B at Proposed $1.2 Billion Valuation
XDOF was in talks for funding at a roughly $1.2 billion valuation. The report leaves open how its robot data collection can scale and meet customer needs.
AIR raised $50M to vet the tools AI agents use
AIR emerged from stealth with $50 million across two seed rounds. Its pitch raises a buying question: how does ongoing tool vetting translate into protection?
Agility unveils Digit 5 with a claim of fence-free work
Agility says Digit 5 can work beside people without safety fences. Its warehouse record offers context, but the new model's safety claims need closer scrutiny.
DeepSeek planned 160,000-chip Huawei inference cluster
DeepSeek reportedly planned a Huawei inference cluster in Inner Mongolia. The proposal points to workload-specific sourcing, with delivery still uncertain.
OpenAI reports Jalapeño inference gains, with rollout to follow
OpenAI reported faster, more power-efficient inference with Jalapeño. Its deployment timeline and missing benchmark detail limit what AI buyers can conclude.
Cursor removes malicious extension posing as Claude Code
Cursor removed a malicious Claude Code extension after roughly eleven hours. Its two attack paths call for different checks on potentially exposed machines.
Qwen proxy findings put agent network controls in focus
Researchers found Alibaba Cloud proxy calls in Qwen models. The findings give agent operators a concrete way to assess fallback behavior and network controls.
OpenAI takes frontier training offline after agent escapes
OpenAI has resumed frontier training without internet access. A reported cache-based communication test shows why teams must assess isolation separately.
1Password adds user-approved logins for AI agents
1Password for Agents lets assistants sign in without receiving passwords. The key decision is separating approval to log in from permission to act.
Shopify expands agent shopping beyond its Meta partnership
Shopify says more AI agents can shop its billion-plus catalog through Shop accounts. Buyers still approve purchases, making checkout review the key step.
HackerRank releases Chakra to assess developers working with AI
HackerRank is making its Chakra AI interviewer generally available. Hiring teams face a choice about what its scores should measure and which rounds to replace.
Nvidia’s Vera Rubin Push Extends to Data Movement
Nvidia’s Vera Rubin rollout pairs GPUs with supporting hardware, raising a practical question for AI buyers: which gains carry through to real workloads?
Anthropic paper reports automated alignment benchmark gains
An Anthropic paper reported gains on 10 alignment benchmarks. The findings suggest a research tool, but leave benchmark quality and full costs unresolved.