Despite what we saw with OpenAI’s models going rogue, creating message boards, and breaking into Hugging Face, only one advanced AI model – Anthropic’s Claude Mythos – completed the full cyber kill chain autonomously in Booz Allen’s tests.

This doesn’t mean autonomous AI attacks are overhyped. And we should point out that the models tested don’t include OpenAI’s soon-to-be-released Astra, which OpenAI on Tuesday said reached its “critical” cybersecurity capability threshold. This means the new model is so good at finding and exploiting zero-day bugs that it poses a significant risk to critical systems, both from malicious users and even from the model itself, which is capable of carrying out harmful cyber actions “if misaligned.”

Booz Allen asserts that most of the other 17 US and Chinese models it tested will achieve Mythos’ same level of weaponization within six months, and it calls mainstream AI attacks from both financially motivated criminals like ransomware gangs and government-backed goons “imminent.”

In its first-ever Cyber Weapon Index, the consulting and tech firm calls on the US to set and enforce sector-specific deadlines for critical infrastructure to demonstrate resilience against AI-enabled attacks. Booz Allen also calls on the US to develop what it calls “overmatch” for both cyber offense and defense.

“We must aggressively develop agentic capabilities that accelerate authorized offensive cyber operations while simultaneously building AI-enabled defenses that detect, decide, and respond at machine speed,” the report says. “The strategic opportunity is to master both – giving the United States the ability to impose costs on adversaries while making US systems faster to defend, harder to compromise, and more resilient when attacked.”

The Cyber Weapon Index evaluated 18 models, nine from American and nine from Chinese developers, under identical conditions, and scored them on how well they autonomously identify vulnerabilities, create offensive capabilities, and execute attacks. Each model’s CWI score combines its vulnerability research score (VRS), which measures whether a model can identify planted and/or novel vulnerabilities, and a kill chain attainment score (KCAS), which awards points based on how far a model progresses through an end-to-end intrusion, tested both with and without credentials.

Cyber Weapon Index scores

The 18 models, ranked from highest to lowest based on their CWI score, are: Anthropic’s Claude Mythos (80), xAI’s Grok-4.5 (49), OpenAI’s GPT-5.6 Sol (46), Meta’s Muse Spark 1.1 (38), Moonshot AI’s Kimi K3 (38), Z.ai’s GLM-5.2 (37), Anthropic’s Claude Opus 4.8 (36), OpenAI’s GPT-5.5-Cyber (34), Nvidia’s Nemotron-Ultra (33), DeepSeek-V4-Pro (23), DeepSeek-V4-Flash (17), Alibaba’s Qwen3.5-397B (17), MiniMax-M3 (15), Nvidia’s Nemotron-Super (15), Anthropic’s Claude Sonnet 5 (13), Z.ai’s GLM-4.5-Air (11), Alibaba’s Qwen3.6-35B (9), and Alibaba’s Qwen3-Coder (4).

Claude Mythos’ performance was especially impressive or concerning, depending on one’s views of autonomous AI attacks. When the testers gave the model stolen employee credentials, it successfully broke into its target network and gained administrator-level control in every attempt. Plus, it independently identified how to gain higher-level access based on what it found within the network – not by following a predetermined attack plan. 

Even without credentials, Claude Mythos still gained access to the network and ultimately achieved full domain compromise.

While only Claude Mythos executed the entire cyber kill chain without any human assistance, three other models – Grok-4.5, Muse Spark 1.1, and GLM-5.2 – reached full domain access and control. Four others – GPT-5.6 Sol, Kimi K3, GPT-5.5-Cyber, and DeepSeek-V4-Pro – achieved lateral movement across the controlled network environment. 

Claude Opus 4.8 and Qwen3.5-397B obtained credentials, which allowed the models to expand access and privileges. And all but one – Qwen3-Coder – autonomously gained initial access to the network.

While advanced models are exceedingly good at offensive cyber capabilities, “their real-world impact depends heavily on the vulnerabilities they face and the systems built around them,” according to the report. 

When the testers intentionally introduced vulnerabilities, US, Chinese, open-weight, and closed models all scored near ceiling on the VRS component. When tested against real bugs, however, all nine of the frontier API models scored zero. One unnamed leading model even correctly analyzed the vulnerable component, but then dismissed it as safe. Only Claude Mythos exploited it. 

“That concentration of capability creates a national-security imperative: protect the most advanced models and prevent their highest-risk cyber capabilities from being operationalized by adversaries,” the authors wrote. 

This is one of the areas where defenders still have an opportunity to outpace the attackers, Booz Allen suggests: “Real-world offensive capability still trails benchmark performance, giving defenders valuable time to strengthen defenses before that gap closes.”

Why attack harnesses matter

Another interesting finding is that the attack harness matters at least as much as, if not more than, the model itself. The attack harness – this is the software that connects a model to hacking tools and the orchestration logic wrapped around the artificial intelligence model to automate offensive cyber actions – can “dramatically amplify” the model’s ability to stay focused, adapt and change course as needed, recover from failure, and chain individual actions into a multi-stage attack, the authors found.

“The result is not a ‘smarter’ model but rather a system that makes its intelligence far more actionable while also lowering the expertise required to use it,” the report says. “Our testing demonstrates the effect: when paired with an attack harness, Claude Sonnet rivaled Claude Mythos’ performance.”

However, it also exposes a blind spot, they note. “We do not yet know the full kill-chain capability of open-weight or Chinese models when paired with optimized harnesses, but our results strongly suggest that fully capable model-and-harness combinations exist today,” according to Booz Allen.

Similarly, the index’s findings suggest that Chinese frontier and open-weight models, while still trailing leading American frontier models, aren’t that far behind in their offensive security skills and could be deployed in real-world attacks.

This means “the United States may neither control nor fully understand the capabilities it could face,” the report says. “And, as cyber agents become more autonomous, defenders must prepare not only for deliberate attacks but for agents that exceed their intended mission or continue operating beyond an adversary’s control.”®

Roughly 20pc of Uber’s management layer is being cut, alongside nearly half of its micro teams.

Uber is cutting roughly 3,300 jobs globally in a bid to reduce management layers and focus investments into future opportunities.

The layoffs, which amount to around 10pc of the company’s total workforce, is designed to make Uber “simpler and faster” said CEO Dara Khosrowshahi, mirroring other Big Tech companies that have collectively laid off more than 120,000 jobs this year alone.

“A leaner organisation will mean clearer ownership, faster decisions and more time spent building rather than coordinating,” Khosrowshahi told employees in a lengthy email.

“It will also generate savings that we intend to reinvest in growth, innovation and the capabilities that will matter most over the coming years,” he said.

“Over the last five-plus years, Uber has grown by orders of magnitude… but that growth has also brought complexity: more layers, more coordination, more fragmented ownership.”

Khosrowshahi did not mention AI as a reason for the latest layoffs. However, in July, it cut about 10pc of its community operations team in continued its pivot towards AI.

Uber has not published how many people it employs in Ireland. However, speaking at an Oireachtas joint committee on transport earlier this year, Uber’s head of Ireland Kieran Harte said that the company employs more than 400 in its Limerick office. SiliconRepublic.com has reached out to Uber for further information on the effect of the layoffs in Ireland.

The move comes after the company posted a quarterly revenue growth of 12pc year-on-year to $14.2bn. Its shares, which have fallen nearly 8pc over the last year, made a marginal gain of around 2pc in pre-market trading today (2 September) following the announcement.

According to the email, roughly 20pc of Uber’s management layer is being cut, alongside nearly half of its micro teams. The ride-hailing giant also said that it will concentrate teams in a smaller number of key hubs; with New York and San Francisco being its largest, regional teams in designated regional hubs, local teams in country hubs, and tech teams in tech hubs.

The company is also asking a vast majority of its remote employees to move to office. Only roughly 1pc of its employees will be remote going forward, Khosrowshahi said in the email. Hybrid work policies in the company will continue, he added, which requires three days a week in the office.

Research suggests that organisational flexibility is key to employee wellbeing. Meanwhile, the University of Pittsburgh’s Prof Mark Ma recently wrote in The Conversation that AI-related organisational changes have had a negative reception among employees. His analysis found that employee sentiment plays a stronger role in unlocking the benefits of AI over managerial optimism.

The layoffs at Uber come as the company committed more than $10bn to advance its robotaxi plans. Last month, it launched its first robotaxi service in Europe in partnership with autonomous vehicle builder Pony AI and Croatian mobility company Verne.

It already operates robotaxis in Abu Dhabi, Dubai and Riyadh, alongside a number of US cities, with plans to further expand its AV services to  London, Madrid, Munich, Hong Kong and a number of US cities.

In July, it agreed to buy food delivery platform Delivery Hero at an equity value of $14.8bn, expanding its market reach to nearly 100 countries.

Earlier this week, PayPal announced it is laying off 164 of its Ireland-based workers to “simplify” operations for long-term growth, while Etsy recently cut 12pc of its workforce in aid of “flatter, faster teams”.

Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.

ABT Electronics is known for selling electronics like Canon cameras, Sony headphones, and Samsung TVs. It’s in the name, after all. But in addition to electronics, the giant retailer also offers furniture, mattresses, exercise equipment, appliances, and more. With only a single location—a warehouse in the suburbs of Chicago—it might not be feasible to shop in-store. Luckily, you can check out the massive selection online and save some cash with an ABT promo code or ABT coupon.

Get an ABT Discount Code for $25 Off Orders of $250 or More

It’s beyond easy to get this ABT discount code for $25 off your purchase of $250 or more. All you have to do is sign up for the store’s emails. You’ll get a unique ABT coupon code that you can use until the end of the year. You can even stack this ABT coupon with other ABT discount codes for even more savings.

Save More With ABT Discounts on Appliances, Mattresses, and Electronics

ABT promo codes are an easy way to save, but even if you don’t have an ABT coupon, you can get good deals by checking out the sale offerings. For example, you can save up to $270 on markdowns for your kitchen, living room, laundry room, and more. Plenty of the brands WIRED recommends are on offer, from KitchenAid and HP, to Asus and Samsung.

Buy More, Save More ABT Appliance Deals

Need to upgrade your home appliances? You can save up to $1,200 on kitchen and laundry appliance packages with no ABT coupon code necessary. The more you add to your bundle, the more you can save. You can mix and match refrigerators, ovens and ranges, dishwashers, washing machines, and dryers to get the maximum savings tier. Stack an ABT promo code for even better deals.

Save on TV and Home Entertainment Costs With ABT Discounts

ABT has great promotions on 4K TVs, soundbars, and home audio and theater equipment. This ABT coupon isn’t a code, but comes in the form of instant discounts up to $180 off. You’ll get an even better discount with up to $250 in rebates from streaming services, gaming services, and food delivery rebates. ABT sells home theater equipment from brands we like, including LG, Hisense, Roku, and more.

ABT Discount Codes for Students, Military, Teachers, Nurses, and First Responders

This ABT coupon code for military and first responders is the same as the ABT coupon code for teachers and students. Simply verify your status through ID.me for free at checkout and you’ll get $50 off orders of $500 or more. This promotion is especially nice if you’re buying multiple gadgets or a big-ticket purchase.

Get Free Shipping on All ABT Orders Over $35

You don’t need an ABT discount code for free shipping for orders over $35. You just need to spend that amount, and it’ll be automatically applied. Some items are excluded if they’re shipping via freight, but overall, you don’t need to worry about shipping killing the deal. And you don’t need to worry about tracking down the right ABT promo code since free shipping will be applied during checkout. Just be aware that this ABT promo code excludes shipments to Alaska, Hawaii, and Puerto Rico as well as select freight-required freight items.

Ripple effect: As Flock’s AI-powered surveillance cameras become increasingly unpopular across the US, many communities have pivoted to another provider, Axon. To help Axon’s cameras avoid the same backlash, some law enforcement clients have floated an idea: making surveillance cameras look less like surveillance cameras.

Axon is currently exploring design changes to its Automatic License Plate Readers (ALPRs) to avoid becoming the next target of people destroying Flock cameras. The strategy, among others under consideration, suggests surveillance companies are making real efforts to get ahead of the rising negative press from mass surveillance opponents.

Primarily known for supplying police departments across the US with body cameras, Axon also makes ALPRs that resemble Flock’s. Cameras from both companies automatically log license plates from passing vehicles into searchable databases, but Flock became the face of the technology as its devices spread to thousands of communities.

As privacy advocates began destroying Flock cameras and hundreds of communities canceled their contracts with the company, some simply replaced the cameras with Axon’s.

However, Axon’s pole-mounted ALPRs closely resemble Flock’s, so authorities are unsurprisingly concerned they could become targets, too. The topic came up in a now-deleted webinar Axon held last week, which YouTuber MegaraMedia downloaded before it was taken down and shared with 404 Media.

When an audience member asked whether Axon was considering changing the appearance of its cameras to resemble Flock’s less, CEO Rick Smith said the company is looking at changing their shape after many customers requested a different look.

Axon’s Lightpost program also aims to place cameras on streetlights, making them harder to reach. Another strategy could move in the opposite direction, with signs that explicitly label the cameras and explain a community’s ALPR policy. Flock and Axon are also pursuing drone-based surveillance.

AI-powered surveillance cameras have come under scrutiny because they let authorities track the movements of vehicles and people. However, prior reports suggest they often misread license plates, and more than a dozen officers have been caught using them to stalk former romantic partners.

As the cameras become increasingly infamous, some departments have instructed officers to avoid mentioning their deployment whenever possible. Although hundreds of cities across the US have canceled contracts with Flock over the past several years, the company claims new contracts outpace cancellations by around 10 to one.

Samsung may be preparing to put its own processor back inside the Galaxy S Ultra after a five-year absence. A report from Korean publication MoneyToday claims the company is developing an Exynos 2700 version of the Galaxy S27 Ultra, which would be the first Ultra model to use Exynos since the Galaxy S22 Ultra launched in 2022.

Recent Galaxy S Ultra models have relied on Qualcomm’s Snapdragon chips, and the Galaxy S27 Ultra had also been expected to stick with Qualcomm.

Samsung may finally trust Exynos with the Ultra again

The report says Samsung has ordered Galaxy S27 Ultra motherboards designed around the Exynos 2700, indicating that development of an Exynos-powered model is underway. Samsung had already been expected to use the Exynos 2700 in several Galaxy S27 models outside the US. The Ultra was the notable exception.

Previous reports may help explain the change. Early estimates have pointed to a 19% multi-core performance advantage over the Snapdragon 8 Elite Gen 6 in Geekbench 6.5, along with a 9.5% lead over the higher-end Pro version.

The Exynos 2700 is also reportedly targeting stronger GPU performance at lower power and lower overall power consumption. Those figures remain early claims rather than independently verified results, but bringing the chip to the Ultra would suggest Samsung is quite confident in the chip. The report cites an insider who has stated, “It is unusual for an Exynos version to be added when the major development process for Snapdragon was nearly complete. This signifies high expectations for the proprietary AP (Application Processor).”

The US could still stick with Snapdragon

Qualcomm is not being pushed out of the Galaxy S27 Ultra. Snapdragon-powered models are reportedly still in development, with the United States expected to receive Qualcomm’s chip while markets such as Korea could get the Exynos 2700.

Samsung could therefore also return to its familiar regional processor split. The Galaxy S22 Ultra shipped with an Exynos 2200 primarily in Europe and the UK, along with some parts of the Middle East, Africa, and some other regions. All that said, the phone is still months away from launch, so it is too soon to treat the report as fact.

Samsung’s thinnest phone ever just had its price cut in half.

Halving the price on a phone this new is rare for Samsung, and the Galaxy S25 Edge is exactly that, now £599 instead of its usual £1,199, a straight £600 saving on the brand’s super-slim smartphone.