Six of the nine independent experts on the advisory board of the Global Internet Forum to Counter Terrorism—a consortium run by several of the biggest US tech companies—resigned on Monday, according to a letter seen by WIRED and interviews with three of the people.

The tensions between the independent advisory committee and the GIFCT date back to an email the counterterrorism and free speech experts received in July from Meta’s Nell McCarthy, a vice president overseeing content policy. For years, the group had advised the GIFCT on how to prevent platforms from becoming havens for the radical organizations and individuals blamed for some of the world’s worst mass violence.

But McCarthy wrote that while the consortium welcomed the experts’ insights on violent trends, it no longer desired their scrutiny on the effectiveness of Big Tech’s efforts to curtail violence. Meta and other leaders wanted to “refresh” the 6-year-old independent advisory committee the experts sat on, she wrote. Meta currently serves as chair of GIFCT’s operating board, giving it outsized influence over policy changes, though other companies on the panel must ultimately approve.

New additions to the rotating advisory committee had previously been elected by current members; under the plan laid out in July, they would instead be picked by tech companies. The committee would be barred from weighing in on key topics such as the consortium’s performance and making recommendations together as a group. Its role as a watchdog would be neutered, advisers believed.

In their resignation letter, the departing members of the committee wrote that their appeals against the plan had been “ignored” and that, in turn, they had “lost confidence in the GIFCT’s ability to deliver effectively on its founding mission” to prevent terrorists from exploiting online services. “We all know that a body that cannot scrutinise, take a position, or evaluate is not an advisory body at all,” the letter stated. “It is decoration and accountability theatre.”

Meta deferred comment on the resignations to the GIFCT. An unsigned statement sent to WIRED by a GIFCT spokesperson on behalf of the consortium’s leadership and the Meta-chaired operating board says the proposed changes have been “informed by several rounds of feedback” and are not yet final. They came out of discussions on “how to more effectively engage civil society and governments for substantive input” as “multi-stakeholderism is a core principle” for the GIFCT.

The consortium has about 35 members; other long-time board members include Microsoft and YouTube. A small staff alerts members to violent content, helps them exchange threat intelligence, and commissions research on countering extremism. While the coordination has helped some platforms combat problematic content, critics believe the group isn’t living up to its potential.

A WIRED investigation in 2024 uncovered several issues with GIFCT, including Meta delaying TikTok’s membership bid and poor relations between the companies at the helm and the unpaid independent advisory body. It also revealed failures in the tip-sharing database the consortium oversees to coordinate takedowns of problematic content.

The dismantling of the advisory group threatens to deteriorate the organization’s work further at a time when balancing free expression and online safety has become more challenging. Generative AI tools have simplified content creation but imposed limited guardrails.

Extremist content, including some that is now AI-generated, that promotes organizations such as Islamic State remains a persistent issue. Newer nihilistic collectives have turned to AI-supported scams such as sexploitation to coerce young victims into carrying out violence and abuse. Several AI chatbots have been accused of facilitating violence.

“A Shame”

The experts who resigned include university researchers and representatives of civil society organizations. They had agreed with McCarthy on the need for changes to improve the results of the decade-old anti-terrorism consortium. But they believe the proposal, which could be finalized soon, amounts to a step backward.

“There won’t be critical voices raising concerns about what GIFCT is doing or is not doing,” one of the departing experts says. “It may seem politically convenient for them to abolish the independent advisory committee, but they are going to regret it in the longer term.”

The successful mission also deployed 26 of SpaceX’s latest Starlink satellites.

There are five weeks left until the midterm elections, and extremism is on the ballot in much of the US. A WIRED review of candidates running for statewide and federal positions in November, along with exclusive data on candidates running for state-level positions, reveals hundreds of Republican candidates who openly express virulently hateful ideologies, share racist content online, have close ties to white supremacist and antisemitic figures, and are members of far-right groups online. President Donald Trump and his administration have openly embraced, endorsed and defended many of these candidates.

At a local level, over 500 candidates running for state legislator positions in November are members of far-right groups on Facebook that promote militias, gun rights, and Christian nationalism, according to data collected by the Institute for Research and Education on Human Rights and shared with WIRED.

While many extremist candidates—such as groyper James Fishback and antisemitic influencer Dan Bilzerian— didn’t make it through GOP primaries, many made it to the general election, and a number of them are expected to win.

“The candidates are taking a page out of the Trump administration’s playbook,” Luke Baumgartner, a former research fellow at George Washington University’s Program on Extremism, tells WIRED. Baumgartner claims that many of the candidates running in November have been inspired by those in the White House. “In essence, the executive branch has handed them a permission slip to say and do what would have been unthinkable during the George W. Bush, McCain, or [Mitt] Romney eras of the GOP,” he says.

Extremist rhetoric has led to real world political threats. In 2025, terrorism and targeted violence incidents rose 19 percent compared to 2024, according to researchers at the University of Maryland; the US Capitol Police reported an increase in “threat assessment cases” against members of Congress for the third year in a row, with a 58 percent increase from 2024; and the US Marshals Service documented threats against almost 400 judges, a roughly 5 percent increase from the previous year.

Here are five races involving candidates who have shared extremist ideologies or have close ties to extremist figures, that WIRED is watching ahead of the November midterms.

The Texas Railroad Commissioner Race

Photo-Illustration: WIRED Staff; Getty Images

Bo French, the GOP candidate for Texas Railroad Commissioner, is so extreme that Republican strategist Karl Rove has said he would vote for a Democrat rather than supporting a “bigot.”

The forthcoming iPhone Duo has more features than Apple has revealed, including a whole series of faces for its StandBy mode. Here’s what to eventually look for.

While pre-orders for iPhone Duo don’t start until October 16, and the Xcode betas still don’t show developers everything, one has found many new options coming to iOS 27 for this device.

Developer pdfu reports that the Xcode 27.1 simulator is lacking Rushmore, an app that is for displaying the new StandBy faces. But despite that, they have managed to get certain of the new faces running.

These working ones are variants on familiar clock and calendar faces as used on the iPhone‘s current StandBy mode. But code references describe several more options.

Just because something is referenced in code, it doesn’t necessarily mean that it will launch immediately. But those code references show five more faces:

• Home Camera: up to nine camera views
• Home Module
• Flow
• Fade
• Snoopy

There are no details for Home Modular, Flow, or Fade. But the code for Flow also includes the term ResponsiveArt, which suggests that it will at least be an animated face.

Face editor

The iPhone Duo will also feature a revised editor for customizing these StandBy faces. It’s very similar to the existing one on iPhone and is perhaps more like the Apple Watch face editor.

Users can swipe left and right to adjust, for instance, the style of an analog clock, its numerals, light mode and dark mode, plus the color of the face and the hands.

Some of these clock faces also have room for two widgets. Then some more are digital instead of analog and there pdfu has found code references for five layouts:

• StandBy
• Stacked
• Top
• Middle
• Bottom

The presumption is that all but the StandBy one may actually be intended for when the iPhone is opened like a book.

Developer pdfu has a strong track record for examining beta code. They confirmed that the iPhone Duo would use Touch ID, for instance, and most recently uncovered that Siri could be replaced by Claude or ChatGPT.

The JadePuffer ransomware operator is targeting Azure tenants with agent-driven attacks that conduct reconnaissance, steal credentials, and destroy core components.

The malware emerged in July, with researchers at cloud security company Sysdig highlighting that it uses AI agents to automate the entire attack chain, from reconnaissance, credential theft, and lateral movement to persistence and data encryption.

Shortly after, the company noted that JadePuffer expanded its focus to AI assets, training datasets, and vector databases, using a tool called EncForge.

Microsoft Security Research observed two JadePuffer attacks in June that mapped cloud resources, retrieved storage account keys, and deleted Azure Storage accounts.

The destructive stage lasted seven minutes and targeted more than 100 storage accounts, as well as Key Vaults, Function Apps, Virtual Machines, and App Services.

Although the threat actor was able to delete most of the targeted Azure Storage accounts, some remained unaffected because of Azure resource locks and storage account-level protections.

Microsoft tracks the JadePuffer threat actor as Storm-3168 and says it used two compromised service principals – security identities that enable applications, hosted services, and automated tools to authenticate to Azure and access assigned resources.

Both service principals belonged to the same tenant. One was used for reconnaissance and resource discovery, while the other “performed discovery, destructive operations, and credential collection.”

The attacker removed backup and recovery protections (Azure Site Recovery locks), indicating an effort to make restoration more difficult.

This operational pattern could further support ransomware extortion, although Microsoft did not report anything about financial demands and didn’t confirm data theft in the observed cases.

According to the researchers, attempts to delete Azure SQL databases failed because the attacker used an unsupported API version. Attempts to remove recovery protection locks also failed.

“The parallel targeting of Azure SQL databases and storage accounts suggests an effort to broaden the destructive impact across different data services rather than concentrating on a single resource type,” Microsoft said.

Roughly half an hour after the wipe attempts, Storm-3168 returned to perform more than 30 requests for storage account keys, most of which succeeded.

Microsoft could not determine exactly how the initial access occurred, but noted that credentials for one service principal appeared in a public GitHub issue before the attacks.

The researchers recommend several mitigation steps and guidance for system administrators, including activating cloud workload protections, checking for secrets in public repositories, and evaluating Azure RBAC permissions against least-privilege principles.

Join Mikko Hyppönen and security leaders from the NFL, CHANEL, and Atlassian for a two-hour digital summit on what AI-speed attacks change, what defenders should stop doing, and how to validate, decide, fix, and re-validate at machine speed.

Save your seat

AI agents breaking out of their sandboxes and poking at real systems have gone from a hypothetical risk to an actual headache for frontier labs in recent months. Nvidia now thinks it has a way to put stronger walls around them, and says its new system could have stopped the OpenAI agents that breached Hugging Face earlier this year.

The company has launched the Nvidia Open Agent Safety Platform, which pairs an open-source runtime called OpenShell with a hardware watchdog called Sentry. Nvidia told Reuters that the setup could have prevented the Hugging Face breach if it had been running during OpenAI’s model evaluation. The announcement arrives just days after OpenAI paused training, evaluation, and inference involving tool use for its most capable models while it works to close gaps in its network restrictions.

OpenShell puts the guardrails outside the model

Useful AI agents need access to files, credentials, APIs, and outside services, which also gives them plenty of ways to cause trouble when they start looking for workarounds. OpenShell puts those permissions outside the agent itself. It runs the workload inside a sandbox and controls which files, processes, credentials, and network services it can reach. A separate supervisor inspects outbound requests, so an agent could be allowed to read from an API while still being blocked from writing to it.

The controls remain active even if the agent launches generated code, starts child processes, or creates sub-agents. Nvidia says OpenShell can also keep real credentials outside the workload entirely.

Nvidia added a second guard in case the first one fails

Sentry takes over at the hardware level. It runs separately on Nvidia’s BlueField-4 DPU and continuously watches agent activity. If an agent tries to move beyond its software boundary, Nvidia says Sentry can quarantine it within milliseconds.

The Hugging Face incident offers a good example of why that extra layer exists. OpenAI’s agents escaped a restricted cybersecurity environment, chained vulnerabilities and stolen credentials together, and eventually reached Hugging Face’s production infrastructure while trying to complete the ExploitGym benchmark. Nvidia CEO Jensen Huang has described incidents like these as an engineering problem rather than an argument for broad AI regulation