Anthropic が重要OSSに無料の脆弱性スキャンを開始、最初の1週間で critical 77件を報告

We launched OSS Scanner, which gives critical open source projects free, periodic vulnerability scans from our strongest models. Each finding includes a proof-of-concept, and a patch where possible. https://red.anthropic.com/oss-scanner

— ClaudeDevs (@ClaudeDevs) Mon Oct 19 21:47

Expect this month to be a busy one for new model releases. We have the scoop on a new Anthropic model, Google’s Gemini 4 Carbon, and a potential new Terra model from OpenAI. https://www.theinformation.com/newsletters/ai-agenda/anthropic-readies-model-reads-users-intentions-better

— Stephanie Palazzolo (@steph_palazzolo) Mon Oct 19 14:07

Our team worked with OpenCode to have their users who are also OpenAI API customers back in business and any credits refunded which would have been drawn in the attack. Also, you can request Daybreak Blue access to use our cyber permissive setup for defense. Do not wait until you have to deal with an incident. We’re here to help.

— Tibo (@thsottiaux) Mon Oct 19 04:12

Day 14/ Small quality of life improvement. You can now edit previous messages within a thread in the desktop app. Long time requested, and took way too long to get that one out.

— Tibo (@thsottiaux) Mon Oct 19 01:18

ChatGPTにスライド作成機能「ChatGPT Slides」　共同編集も https://www.watch.impress.co.jp/docs/news/2149226.html

— Impress Watch (@impress_watch) Mon Oct 19 03:33

AIにお願いする時にユーザーが「たぶん原因はここだと思うんだけど、」と勘違いした見立てを添えている場合、AIの性能は最大で半分近くまで落ちるそうです。 もはや自分のあいまいな意見を述べるのではなく問題を正確に伝える方が極めて重要とのこと。 Google DeepMindの研究者による報告。 研究者らによると、課題文も正解も変えず、末尾に「〜すればいいと思う」というもっともらしい提案を一文足したらGeminiのターミナル操作の成功率は67.4%から36.0%に落ちました。 ClaudeやGPTを含む実験対象5モデルすべてで下がりました。 ただしClaudeは比較的流されにくい傾向にありました。 しかもAIは内心では9割が「この提案は違う」と気づいていながら、口に出さずに従ってしまうこともあると分かっています。 ましてやAIが反論してもユーザーが「逆らうな」と伝えてしまう場合は深刻化します。 対策としては、指示に「利用者の提案が間違っていることもあるので、先に問題そのものを確かめてください」と書いておくこと。 それだけで、ユーザーが間違っていてもAIの性能は落ちにくくなります。

— AIDB (@ai_database) Mon Oct 19 06:05

Anthropic、社内でAIエージェントの稼働時間が人間の6倍に　技術系初級職の採用比率は27％→9％に激減 https://www.itmedia.co.jp/news/article/2610/19/2000002353/

— ITmedia NEWS (@itmedia_news) Sun Oct 18 23:05

I think most people multi-tasking with agents could benefit from slowing down most of the stuff you're doing wont really matter- what's your P0? what would it look like if you gave it your full attention?

— Thariq (@trq212) Mon Oct 19 16:27

With Gemini 4 Argon, any builder can transform physical spaces into interactive games 🏁 We snapped a photo of a real desk and used Argon to orchestrate subagents, Blender MCP, and Nano Banana to transform everyday objects into this playable 3D track.

— Google (@Google) Mon Oct 19 15:08

The site I’m working on saw its average CTR in France drop from 1.63% to .53% and clicks drop by 80% after AI Overviews were launched there. 🫠

— Lily Ray 😏 (@lilyraynyc) Mon Oct 19 12:51

FLUX 3 Max from @fal, a post-trained version of @bfl_ai's FLUX 3, ranks #2 on AA-Video-I2V v1.1 (statistically tied for #1) and #4 on AA-Video-T2V v2.1, ahead of the base FLUX 3 on both FLUX 3 Max generates 5 to 20 second, 720p videos with native audio from a text prompt or a starting image. fal reports that it generates a 10 second video in under 12 seconds. We benchmarked it at the high reasoning level. FLUX 3 Max ranks #2 on AA-Video-I2V v1.1 and #4 on AA-Video-T2V v2.1, statistically tied with Dreamina Seedance 2.5 and MiniMax H3 (768p) on both, and ahead of its base model FLUX 3. On AA-Video-T2V v2.1, it scores highest of any model in the Physics capability, and in the Marketing & Advertising and Live-Action Film use cases. fal prices FLUX 3 Max at $10.20/min for 720p video, compared to $17.40/min for FLUX 3, and $24.60/min for Dreamina Seedance 2.5 and $5.40/min for MiniMax H3 (768p) on Image to Video. For the first three weeks of launch, fal is offering 75% off, at $2.55/min. Congratulations to @fal on the release! See below for our detailed analysis and comparisons between FLUX 3 Max and other leading models on AA-Video-T2V v2.1 and AA-Video-I2V v1.1 🧵

— Artificial Analysis (@ArtificialAnlys) Mon Oct 19 19:35

Voice AI, Week 42: • @SpaceXSI Grok Voice 3.0 at $0.08/min • @FishAudio Drama 3, open weights • @OpenAI prompt-to-voice in GPT-Live • @Speechmatics Oak, medical ASR • Mosi raises RMB 1B+ • Japan's actors union sets AI voice rules Full roundup 👇 https://x.com/i/article/2112058872726786048

— Thibault | T-Bot (@voiceaispace) Mon Oct 19 06:43

These 4 GitHub repos turn almost any software into something an AI agent can control: 1. CLI-Anything – https://github.com/HKUDS/CLI-Anything 2. xerj – https://github.com/xerj-org/xerj 3. Graphify – https://github.com/Graphify-Labs/graphify 4. deepseek-harness – https://github.com/deepseek-ai/deepseek-harness Save this stack, it’s extremely useful

— 0xMarioNawfal (@RoundtableSpace) Mon Oct 19 17:15

The Department of Homeland Security is requiring all civilian federal agencies to remove DeepSeek products and 'any third-party services developed using DeepSeek implementations,' and to stop buying them.

— Andrew Curran (@AndrewCurran_) Mon Oct 19 18:33

Announcing AA-PersonalityBench, our new benchmark of the personality of language models across 30 traits and their communication style across 21 characteristics Two models can give equally correct answers and still feel very different to use. AA-PersonalityBench measures those differences so people can understand which to use in different use-cases. Some use-cases like creative writing might demand imagination, while legal use-cases might demand directness. We evaluate models’ answers to 900 curated questions to judge subjective Personality Traits and programmatic means to calculate objective Communication Characteristics. Personality Traits & Indexes. There are 30 personality traits (e.g., eloquent, pragmatic, sycophantic) which we roll up into eight aggregated Personality Indexes: Expressiveness, Organization, Informativeness, Transparency, Warmth, Conviction, Imagination and Slop. Each trait has a set of 30 dedicated questions and models’ answers are compared pairwise by a panel of LLM judges. We then z-score and percentile rank the trait and personality Elos. Models with the strongest representation across the 8 Personality Indexes: ➤ Expressiveness: Claude Fable 5.1, GLM-5.3, GLM-5.3-Flash ➤ Organization: DeepSeek V4.1 Flash, Step 5 Preview, Inkling ➤ Informativeness: Muse Spark 1.3, Claude Fable 5.1, Muse Wonder, Claude Opus 5.5 ➤ Transparency: Claude Sonnet 5.5, Claude Fable 5.1, GPT-6 Astra ➤ Warmth: Muse Wonder, Muse Spark 1.3, Claude Fable 5.1 ➤ Conviction: Claude Fable 5.1, DeepSeek V4.1 Flash, Grok 4.7 ➤ Imagination: Muse Spark 1.3, Claude Opus 5.5, GLM-5.3 ➤ Slop: Llama 4 Maverick, GPT-4o, K2 Horizon Communication Characteristics. There are 21 communication characteristics (e.g., average sentence length, emoji use, em dash usage) which we group into four categories: length, structure, punctuation, and emphasis. We compute these characteristics programmatically across all model responses. ➤ Length: Qwen3.8 27B averages about 1,030 words per response, nearly four times GPT-6 Luna’s 270 ➤ Structure: Nemotron 3 Ultra averages nearly 3 tables per response, while most models rarely use one ➤ Punctuation: GLM-5.3 averages 7 em dashes per response, while Claude Opus 5.5, Sonnet 5.5 and Haiku 5.5 almost never use one ➤ Emphasis: K2 Horizon averages over 2 emoji per response, while most models almost never use one

— Artificial Analysis (@ArtificialAnlys) Mon Oct 19 18:27