Edition, Fri, 18 Sep 2026.

Anthropic nears $2 trillion IPO as CEO warns of AI's "real dangers"

The Claude maker is preparing for a potentially record-breaking public offering valued at roughly $2 trillion while its CEO acknowledges the technology carries existential risks.

The Claude maker is preparing for a potentially record-breaking public offering valued at roughly $2 trillion while its CEO acknowledges the technology carries existential risks.

Anthropic is reportedly eyeing share sales that would make it one of the largest IPOs in history, surpassing SpaceX's $1.77 trillion listing earlier this year. The company has posted back-to-back profitable quarters, demonstrating strong commercial traction alongside its safety messaging. Amodei wrote in a recent essay that he will not lie about the dangers of AI. The IPO announcement comes amid growing scrutiny over the role of AI in military applications, though Anthropic has historically positioned itself against certain military and surveillance uses of its technology. Analysts note that Amodei's warnings may actually reinforce the power of Anthropic's technology in the eyes of investors, rather than diminish its value.

Google DeepMind launches institute to shape the debate over AGI safety

The new institute publishes its first essays on frontier model oversight, reasoning transparency and the case for a U.S.-led standards body.

The new institute publishes its first essays on frontier model oversight, reasoning transparency and the case for a U.S.-led standards body.

Google and Google DeepMind launched the DeepMind Institute to advance conversation around artificial general intelligence, listing DeepMind co-founder Shane Legg, Google executive James Manyika and DeepMind chair Demis Hassabis as directors. The inaugural collection of four essays addresses economic policy for AGI disruption, human-readable model reasoning, principles for human flourishing, and a framework for evaluating frontier models. In one essay, Hassabis proposes a U.S.-led standards body that would initially evaluate models on a voluntary basis before potentially requiring deployment approval, with "held-out" independent tests developed by the body. Another essay by DeepMind safety researchers Rohin Shah and Anca Dragan argues that AI's shrinking transparency window is not inevitable and urges developers and regulators to confront the safety trade-offs directly.

AI-enabled bioweapons threat prompts calls for urgent biotech safeguards

Researchers warn that the convergence of advanced language models and accessible biotechnology creates unprecedented risks for designing and deploying biological weapons.

Researchers warn that the convergence of advanced language models and accessible biotechnology creates unprecedented risks for designing and deploying biological weapons.

MIT Technology Review reports that AI companies have issued urgent warnings about the intersection of frontier AI and synthetic biology. A 2022 study using an AI drug-discovery model generated 40,000 molecules with potential as chemical warfare agents in under six hours, some reportedly more toxic than known nerve agents. Anthropic's own September report acknowledged attempts to use its models to make the chikungunya virus more transmissible, create a more dangerous form of bird flu for humans, and build an "atlas of venom toxin peptides." Experts like MIT biologist Kevin Esvelt argue that given the existential stakes, the scientific community must err on the side of caution. While some biologists contend current AI is not yet sophisticated enough for full bioweapon development, the consensus among safety researchers is that the threat is imminent and demands better surveillance, screening tools and precautionary measures.

US AI labs back "superintelligence slowdown" as Europe pushes back

Anthropic, OpenAI, xAI and Google DeepMind have publicly endorsed proposals to pace the frontier of AI development, but European firms and experts argue the plan is a self-serving attempt by US incumbents to entrench their dominance.

Anthropic, OpenAI, xAI and Google DeepMind have publicly endorsed proposals to pace the frontier of AI development, but European firms and experts argue the plan is a self-serving attempt by US incumbents to entrench their dominance.

Anthropic CEO Dario Amodei launched the "pace the frontier" movement with an essay proposing three steps: embedding independent third-party auditors inside frontier labs, coordinating industry-wide safety standards among democratic countries, and seeking global coordination on pacing limits. OpenAI chief Sam Altman and xAI founder Elon Musk both agreed with the slowdown concept, while Nvidia CEO Jensen Huang pushed back, arguing capitalism's market forces are sufficient and calling new regulations "completely unnecessary." Google DeepMind's Demis Hassabis proposed a U.S.-led standards body with independent "held-out" evaluations.

European companies rejected the slowdown proposal. Mistral warned that incumbents are "using this moment to consolidate their market position," while Hugging Face CEO Clement Delangue said "it's not time to slow down but to accelerate." French Finance Minister Roland Lescure called the push "making everyone behind them slow down so they can stay in first place." King Charles III also convened Amodei, Hassabis and Nvidia's Jensen Huang at a Scottish summit to discuss the issue.

Crusoe raises $3.9 billion for modular AI factories and data center expansion

The AI infrastructure company's Series F round values the firm at $30.9 billion and funds deployable Spark factories alongside major data center projects.

The AI infrastructure company's Series F round values the firm at $30.9 billion and funds deployable Spark factories alongside major data center projects.

Crusoe's round was co-led by Atreides Management, Mubadala Capital and Valor Equity Partners, with participation from Founders Fund, GIC, Nvidia, Qatar Investment Authority, Radical Ventures and TPG. The company announced three new board members, including Cloudflare CFO Thomas Seifert and Redwood Materials CEO JB Straubel. Crusoe plans to use the capital to finance existing data center projects, including a large site in Abilene, Texas used by OpenAI, as well as smaller modular AI factories called Spark that can be trucked to large power sources almost anywhere. The company recently signed a roughly $13 billion, five-year cloud contract with quantitative trading firm Jane Street and has met with investment bankers about a potential IPO.

ServiceNow's September store release puts Build Agent at the center of the platform

Auto upgrades, the IDE folded into Studio, and a Brazil EA on September 24 mark the biggest Build Agent push yet.

Auto upgrades, the IDE folded into Studio, and a Brazil EA on September 24 mark the biggest Build Agent push yet.

The September 2026 store release is live on Zurich (Patch 12+) and Australia (Patch 5+), and it marks a shift in how Build Agent is delivered. Instances running Build Agent on qualifying patches now get new versions automatically when they are published to the ServiceNow Store, replacing the manual monthly install (a new version can take up to a day to land). The release also folds the standalone ServiceNow IDE into Studio, so pro-code source files and low-code metadata editors live in one environment, Build Agent history persists across conversations, and the agent runs faster in the unified surface.

Build Agent's playbook generation gets six upgrades: the whole playbook (process definition, stages, activities, wiring) now lands in a single XML update set file, runtime permissions are generated at both playbook and stage level, nested playbooks are supported via a Set Playbook Outputs activity, agentic fields can be configured on form and record-based activities (AI Agent plugin required), on-demand launchers are generated, and activities can be marked optional. Also in the release are three new metadata types (Service Catalog dependent questions, transition conditions, and UI styles), parameterized ATF tests and test suites, Box joining the MCP servers Build Agent can connect to, and two new models: Azure OpenAI GPT 5.6 Sol and Anthropic Claude Opus 5 on AWS. Brazil Early Availability, deferred from its original date, is set for September 24, 2026, when Build Agent (Trial) becomes preinstalled on Brazil instances.

AI hit its ROI targets, but ITSM workloads got heavier

SolarWinds' new survey of 844 IT professionals finds 84% say AI paid off, yet 52% report an increased overall workload.

SolarWinds' new survey of 844 IT professionals finds 84% say AI paid off, yet 52% report an increased overall workload.

SolarWinds' 2026 State of ITSM report, based on 844 IT professionals across North America, EMEA, and APAC, confirms the ROI case: 84% said AI met or exceeded expectations, saving close to nine hours per week in aggregate (3.3 hours detecting and flagging issues, 3.0 hours responding to end-user requests, and 2.9 hours triaging tickets), and 86% reported improved employee productivity.

Yet 52% of respondents said overall workload increased, 19% said it stayed the same, and 30% said it decreased. The added work concentrates in managing and maintaining AI tools and integrations (48%), reviewing and validating AI-generated outputs (47%), and training and fine-tuning models (37%). Only 7% encountered no unexpected costs, with staff training and upskilling (48%), data quality and cleanup (47%), and ongoing tuning (45%) the top items, and 83% now spend at least three hours a week on AI monitoring and maintenance. The report finds teams that lean on activity-oriented measures such as ticket deflection are 2.4 times more likely to report increased workload than those using outcome-oriented measures, which only 21% of organizations currently use. Sean Sebring, SolarWinds' ITSM Solutions Engineering Manager, says the difference between success and failure comes down to the data foundation and integration complexity, and advises teams to optimize first, then automate.

The 320-day moving average bear signal is flashing on ServiceNow again

A setup that has been right every time over the past decade is back in play as NOW stalls near $139.

A setup that has been right every time over the past decade is back in play as NOW stalls near $139.

After rallying from its April lows, ServiceNow shares are running out of steam near a declining 320-day moving average, trading modestly lower at $138.59 on September 17. Per Schaeffer's Senior Quantitative Analyst Rocky White, the stock is now trading within 0.75 times the 320-day moving average's 20-day average true range, after spending at least 80% of the prior two weeks and 80% of the previous 42 trading sessions below that trendline.

That same setup has occurred three other times over the last decade, and each time the stock was lower one month later, averaging a 6.9% decline. The near-term backdrop is still firmly bullish, with 40 of 45 brokerages in coverage rating NOW a buy or strong buy versus three holds and two sells, and Schaeffer's Volatility Scorecard sitting at 79 of 100, suggesting the stock has tended to outperform options traders' volatility expectations over the past year.