Microsoft, Google, OpenAI and Anthropic have agreed to support a single shared opt-out signal for AI training in robots.txt. Website owners can use the token __usage_ai-training to block all four companies from scraping content for model training with one line. Anthropic originated the proposal, and Microsoft will begin honoring it on February 1, 2027, while Google plans to follow in early 2027. The signal only applies to future training and does not affect regular search rankings or create licensing agreements.
Microsoft is also rolling out more granular controls through Bing Webmaster Tools starting December 15, 2027. Site owners will be able to choose between blocking AI training entirely, blocking AI grounding, or allowing limited grounding for Microsoft products only. Google separately announced on October 30, 2026 that it plans to support the shared token while maintaining its existing company-specific rules like Google-Extended.
Meta has not yet agreed to the shared signal, according to reports. The group submitted a draft standard to the Internet Engineering Task Force to formalize the approach. OpenAI aims to implement support in the coming months. The signal represents a preference rather than a crawling order, and it does not prevent past scraping that has already occurred.
In separate AI-related news, Anthropic flagged privacy concerns as a major regulatory risk in its S-1 filing ahead of its IPO on October 31, 2026. The company acknowledged that privacy and safety in frontier AI may sometimes conflict directly. Anthropic also stated it cannot guarantee that its internet data scraping for training fully complies with privacy and intellectual property law.
Reddit shares plunged 29.22 percent in a single day, erasing about $7 billion in market value. CEO Steve Huffman attributed the decline to sharp drops in search engine visits during the third quarter, accelerating in October. Google AI Overviews and AI Mode are seen as the primary cause, as they answer queries without directing users to source websites. Reddit currently has roughly 130 million daily unique users.
Etsy updated its AI policies after user complaints about AI-generated products on the platform. Sellers must now include at least one unedited photo of their item, and AI-generated functional guides like crochet patterns are banned. A new reporting option for misleading AI images has been added, though enforcement will not be immediate. Some sellers say the rules still fall short.
North Korean hackers linked to the TraderTraitor group allegedly used AI tools to steal $6.3 billion from Binance. The attack may have started with a compromised employee laptop, and hackers drained accounts in about 20 minutes. Binance said it is setting aside over $6 billion to cover customer losses. The U.S. Secret Service is working to trace and recover the funds, and blockchain firm TRM Labs attributes the attack to North Korea with high confidence.
Security researcher Eric Schiffer warned that the Binance heist could trigger a wave of more advanced AI-powered cyberattacks. Meanwhile, a new malware called VOLTA uses AI to rank stolen victim assets by monetization value. After harvesting credentials and crypto wallets, it sends data to a language model via OpenRouter to assess the haul, falling back to multiple models if the first refuses.
ONLYOFFICE launched version 10.0 of its open-source office suite on October 31, 2026, adding context-aware AI tools across document, spreadsheet, presentation and PDF editors. The release includes four free modular apps and fixes 35 vulnerabilities. Israeli AI entrepreneurs Or Shani and Tomer Naveh, formerly of Jigso, joined Placer.ai with a team of machine learning specialists to help expand the company's AI capabilities.
IMEC tested open-weight models including Qwen, DeepSeek and GLM across over a thousand coding tasks to determine when owning models is cheaper than paying API costs. Smaller models solved between 32.5 and 40 percent of tasks, and GLM reached the level of Opus. The findings suggest that owning models can be more cost-effective for certain workloads.
Key Takeaways
- <ul><li>Microsoft, Google, OpenAI and Anthropic agreed to honor a single robots.txt token, __usage_ai-training, to let site owners block AI training from all four companies at once.</li><li>Microsoft will start honoring the shared opt-out signal on February 1, 2027; Google plans to follow in early 2027; OpenAI aims to implement it in the coming months.</li><li>Meta has not yet agreed to the shared signal, and the rule only applies to future training, not past scraping.</li><li>Microsoft will launch Bing AI training and grounding controls through Bing Webmaster Tools on December 15, 2026, with enforcement starting February 1, 2027.</li><li>Anthropic identified privacy risks as a major regulatory concern in its S-1 filing ahead of its IPO on October 31, 2026.</li><li>Reddit shares fell 29.22 percent, wiping about $7 billion in market value, as Google AI Overviews and AI Mode reduced search traffic to the platform.</li><li>Etsy now requires sellers to include at least one unedited photo and bans AI-generated functional guides, though some sellers say the rules are insufficient.</li><li>North Korean hackers allegedly used AI tools to steal $6.3 billion from Binance, with the U.S. Secret Service working to recover the funds.</li><li>VOLTA malware uses AI via OpenRouter to rank stolen victim assets by monetization value, representing a new use of AI in cybercrime.</li><li>IMEC testing found that owning smaller open-weight models like DeepSeek and GLM can be more cost-effective than paying API bills for certain coding tasks.</li></ul>
Microsoft Bing adds new AI training and grounding controls for site owners
Microsoft announced new Bing AI content controls that give website owners three options. They can block AI training, block AI grounding, or allow limited grounding for Microsoft products only. The controls roll out through Bing Webmaster Tools starting December 15, 2026. Microsoft will begin honoring the new preferences on February 1, 2027.
Four major AI companies agree on one shared opt-out signal for training
Google, Microsoft, OpenAI and Anthropic agreed to read the same robots.txt signal that tells AI crawlers not to train on a website. Until now, site owners had to write separate rules for each company. Microsoft will start honoring the signal on February 1, 2027, and Google plans to follow in early 2027. Regular search rankings are not affected by the change.
Google joins OpenAI and Anthropic with new AI scraping controls
Google, OpenAI and Anthropic are backing a new robots.txt directive that blocks AI training scrapers. Website owners can add a single user agent called __usage_ai-training to stop crawlers from collecting training data. The directive can also allow specific folders like a press kit to remain open. The new signal does not stop past scraping or affect regular search results.
How to block AI training on your website using robots.txt
Website owners can soon block AI training with one entry in their robots.txt file. They add User-Agent __usage_ai-training and Disallow slash to stop all training access. Microsoft, OpenAI, Google and Anthropic support the shared signal. Microsoft will start honoring it on February 1, 2027, and will stop reading older NoArchive and NoCache tags after March 1, 2027.
Google, Anthropic and others back new AI crawling opt-out standard
Google, Microsoft, Anthropic and OpenAI submitted a draft standard to the Internet Engineering Task Force. The new standard lets website operators block all AI training scrapers with one robots.txt entry instead of many. Microsoft also plans to add an AI Grounding setting to control how AI answers access web pages. OpenAI and Google plan to support the standard in the coming months and next year respectively.
One robots.txt line can opt sites out of AI training at Google and OpenAI
A new shared rule in robots.txt could let website owners block AI training from Google, OpenAI, Microsoft and Anthropic with one line. The idea comes from Kevin Kelley of Anthropic and was submitted to the IETF. It uses a simple token called __usage_ai-training. Google, Microsoft and OpenAI have promised to respect it, but Meta has not yet agreed. The rule only applies to future training and does not affect search results.
Google Plans Support for Shared Robots.txt AI Training Control
Google announced on October 30, 2026 that it plans to support a shared AI training control in robots.txt. Website owners could use one token to set training preferences across multiple AI providers. If a site opts out and does not mention Google-Extended, Google will treat it as a block for Gemini training. Google aims to support the token in 2027 and will share progress updates. The plan works alongside existing company-specific rules like Google-Extended.
AI training opt-out: publishers should not have to choose
Google, Microsoft, OpenAI and Anthropic will honor one shared AI training opt-out signal in robots.txt. Publishers can block AI training with a single rule using the token __usage_ai-training. Anthropic is already working on support, OpenAI aims to implement it in the coming months and Microsoft will start on February 1, 2027. The signal states a preference about content use, not a crawling order. It does not create licensing or payment deals.
How North Korean hackers pulled off $6.3B Binance heist
North Korean hackers may have used a compromised Binance employee laptop and AI tools to steal $6.3 billion in crypto. The attack is linked to TraderTraitor, a North Korean cybercrime group. Hackers drained accounts in about 20 minutes and moved funds across many blockchains. Binance said it is setting over $6 billion aside to cover customer losses and expects withdrawals to resume within two days. The U.S. Secret Service is working to trace and recover the stolen funds.
Expert warns of AI hacking bloodbath after Binance theft
Eric Schiffer warned that the $6.3 billion Binance hack could signal a wave of more advanced AI-powered cyberattacks. He said businesses must take cybersecurity more seriously. Binance founder Changpeng Zhao said the attack appeared to be a hack. The U.S. Secret Service is working with partners to freeze and recover stolen funds. Blockchain firm TRM Labs attributes the attack to North Korea's TraderTraitor group with high confidence.
Fake Capital Gazette image removed from restaurant
A fake version of the Capital Gazette was posted and then removed from a restaurant run by a County Council candidate. The article discusses an altered image compared with the real publication. The story was published on October 30, 2026.
Jigso AI founders join Placer.ai to expand AI tools
Israeli AI entrepreneurs Or Shani and Tomer Naveh, co-founders of Jigso, are joining Placer.ai with a team of AI and machine learning specialists. They bring 16 years of experience in AI and machine learning products. Placer.ai helps businesses understand foot traffic and activity at physical locations. The company wants to move beyond traditional analytics toward automated insights. No financial details or product plans were announced.
Etsy updates AI rules but crafters stay doubtful
Etsy changed its AI policies after users complained about AI-generated products. Sellers must now include at least one unedited photo of their item. AI-generated functional guides like crochet patterns are banned. Some sellers, including crochet artist Lucy Cooper, feel the rules do not go far enough. A new reporting option for misleading AI images has been added but will not lead to immediate enforcement.
IMEC tests when owning AI models saves money
IMEC tested open-weight AI models to see when owning them is cheaper than paying API bills. The team ran over a thousand coding tasks using models like Qwen, DeepSeek, and GLM. Smaller models solved between 32.5 and 40 percent of tasks. GLM reached the level of Opus. The study shows that owning models can be more cost-effective for certain tasks.
Reddit shares drop 29% as Google AI takes traffic
Reddit shares fell 29.22 percent in one day, dropping from $129.60 to $91.73. The company lost about $7 billion in market value. CEO Steve Huffman said search engine visits fell sharply in the third quarter and accelerated in October. The decline is linked to Google AI Overviews and AI Mode, which answer questions without sending users to websites. Reddit currently has about 130 million daily unique users.
ONLYOFFICE 10.0 launches with AI tools and stronger security
ONLYOFFICE released version 10.0 of its open-source office suite on October 31, 2026. The update adds context-aware AI tools across document, spreadsheet, presentation, and PDF editors. It also introduces four free modular Apps for tasks like data collection and secure collaboration. The release fixes 35 vulnerabilities and includes about 50 new features and over 1,000 fixes across online, desktop, and mobile editors.
Anthropic flags privacy risks ahead of IPO
Anthropic identified privacy concerns as a major regulatory risk in its S-1 filing ahead of its initial public offering. The company warned that privacy and safety in frontier AI may sometimes be in direct conflict. Anthropic also stated it cannot be sure whether its internet data scraping for model training will comply with privacy and intellectual property law. The filing on October 31, 2026 describes these issues as pervasive legal risks.
VOLTA malware uses AI to rank stolen victim assets
VOLTA is a mature infostealer malware that adds an optional AI layer to rank victims' stolen assets by monetization value. After harvesting credentials, sessions, cryptocurrency wallets, and other data, it sends a structured evidence bundle to a language model via OpenRouter. If the model refuses the assessment, VOLTA falls back to a chain of three models. The malware is self-deleting and uses no attacker-owned infrastructure.
Sources
- Microsoft Bing New AI New Controls With AI Training From AI Search Grounding
- Google, Microsoft, OpenAI and Anthropic back one AI training opt-out
- Google Joins OpenAI And Anthropic In Promising New Scraping Controls
- How to block AI training on your website with robots.txt
- Google, Anthropic and others to support new AI crawling opt-out standard
- One robots.txt line will opt your site out of AI training at Google and OpenAI
- Google Plans Support for Shared Robots.txt AI Training Control
- AI training opt-out: publishers ‘should not have to choose’
- How North Korean hackers pulled off $6.3B Binance heist -- inside job, AI-powered techniques could have been at work
- Expert warns of 'AI hacking bloodbath' following massive Binance cryptocurrency theft
- Fake version of Capital Gazette posted, removed by County Council candidate’s restaurant
- Israeli AI founders bring Jigso team to Placer.ai in push beyond traditional analytics
- Etsy cracks down on AI-copying issues but some users still sceptical
- IMEC shows when owning AI models beats API bills
- Reddit shares have fallen by 29%: Google AI is siphoning off traffic
- ONLYOFFICE 10.0 Brings AI Tools, Modular Apps, and Stronger Security to Its Open-Source Office Suite
- Anthropic details privacy concerns as pervasive regulator risk in advance of IPO | MLex | Specialist news and analysis on legal risk and regulation
- Appraisal by AI: Model-driven victim prioritization in the VOLTA infostealer
Comments
Please log in to post a comment.