Connect with us

Fintech

Anthropic Pushes Back as Hackers Press AI Weak Spots | PYMNTS.com

Published

on

Anthropic published research Monday (Nov. 24) showing its Claude Opus 4.5 model, released the same day, reduced successful prompt injection attacks to 1% in browser-based operations, down from earlier versions that faced higher breach rates when adversaries embedded malicious instructions in web content.

The results show progress in making agent systems more resilient. However, the underlying weakness still exists as browser-based automation becomes more common.

Prompt injection attacks exploit how AI models process instructions. When an agent browses the web or reads emails, attackers can embed hidden commands that direct the model to leak data, forward confidential communications or take unauthorized actions. PYMNTS Intelligence found 98% of business leaders remain unwilling to grant AI agents action-level access to core systems, with trust emerging as the primary constraint on adoption.

The challenge has drawn acknowledgment across the industry. OpenAI called prompt injection a “frontier security challenge” requiring ongoing work. Microsoft ranked it as the top entry in the OWASP Top 10 for large language model applications in 2025. Security researchers highlight that the issue is particularly challenging. It arises from how AI systems handle natural language, not from typical software flaws.

Attack Surface Expands With Browser Agents

Browser use creates distinct exposure. Every webpage and embedded document becomes a potential vector. Security researchers at Brave demonstrated that attackers can embed nearly invisible commands in screenshots that bypass text-based filters.

Security firm AppOmni revealed that ServiceNow’s Now Assist agents could be manipulated to recruit more powerful agents that read or modify records and send emails while built-in protections remained enabled. Research from Smart Labs AI showed agents can be coerced into leaking internal documents during routine tasks, with success rates varying across implementations.

Advertisement: Scroll to Continue

A Fortune 500 financial services firm found that its customer service agent was leaking account data for weeks through a prompt injection attack, resulting in millions of dollars in regulatory fines, according to a blog post by Obsidian.

Training and Classifiers Form Dual Defense

Anthropic’s improvements center on two approaches. The company applied reinforcement learning during model training, exposing Claude to prompt injections in simulated web content and rewarding the model when it correctly identifies and refuses malicious instructions. This builds robustness directly into capabilities rather than relying solely on external filters.

The second layer involves classifiers that scan untrusted content entering the model’s context window, detecting adversarial commands hidden in text, images or interface elements. Anthropic improved the classifiers and intervention mechanisms since the browser extension launched in research preview.

The company also conducts expert human red teaming and participates in external arena-style challenges that benchmark robustness across the industry.

The 1% attack success rate reflects testing against an adaptive adversary combining multiple known techniques. The figure represents meaningful risk rather than a solved problem.

Industry Adopts Layered Mitigation Strategies

Other AI providers have outlined similar defense frameworks combining preventative controls, detection tools and impact mitigation. Microsoft uses hardened system prompts and a technique called spotlighting to isolate untrusted inputs, alongside Prompt Shields integrated with Defender for Cloud. The company developed FIDES, an approach using information flow control to deterministically prevent indirect prompt injection in agent systems.

Google announced autonomous systems that detect and respond to threats in real time, often without human intervention, as part of a broader shift toward AI-driven preemptive cyber defense.

Security experts say the models remain only as reliable as the data feeding them, with accuracy and accountability determining whether prevention at this scale proves economically viable.

The broader consensus across security teams is that no single technique closes the gap. Providers are layering training, classifiers, monitoring tools and internal guardrails to shrink the window in which prompt injection succeeds.

For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.

Continue Reading
Click to comment

Leave a Reply

Your email address will not be published. Required fields are marked *

Fintech

Speed Raises $8 Million to Expand Bitcoin and Stablecoin Payment Solutions | PYMNTS.com

Published

on

The company will use the new funding to build capacity, expand to new regions, develop more merchant tools, enable cross-border and creator payouts, and maintain reliability and compliance, it said in a Tuesday (Dec. 16) blog post.

Speed’s offerings include a global payment layer called Speed Merchant that is designed for merchants, platforms and payment systems and enables them to accept both Bitcoin and stablecoins, according to the post.

The company also offers a Lightning wallet called Speed Wallet that serves individuals and businesses and enables Bitcoin and stablecoin transfers, supports global payouts, offers local on- and off-ramps, and powers USDT transactions, the post said.

“We’ve always believed that Bitcoin and stablecoins can power everyday payments,” Speed CEO Niraj Patel said in the post. “That requires real infrastructure—fast, compliant and scalable. This investment validates that belief and accelerates our mission.”

Speed co-founder Jayneel Patel said in the post that the company aims to “solve real problems with technology.”

Advertisement: Scroll to Continue

“Speed started as a merchant solution and has grown into a global payment network,” Jayneel Patel said, adding the company is “ready to take the next leap.”

Stablecoin issuer Tether and venture fund ego death capital co-led the funding round, per the post.

Tether said in a Tuesday press release that its investment supports its strategy to support Bitcoin-aligned financial infrastructure and expand the utility of its USDT stablecoin in real-world payment environments.

“We support teams building practical infrastructure that reduces friction in payments and expands access to reliable settlement rails,” Tether CEO Paolo Ardoino said in the release.

Tether’s USDT stablecoin is the most traded cryptocurrency by volume around the world.

Adam Gebner, associate at ego death capital, said in a Tuesday blog post that Speed processed over $1.5 billion in payment volume over the past 12 months and serves more than 1.2 million users.

“By bridging Lightning and stablecoins in a single, compliant platform, Speed is positioning itself as foundational infrastructure for the Bitcoin and stablecoin economy, serving merchants, platforms and users across both developed and emerging markets,” Gebner said in the post.

Continue Reading

Fintech

Databricks Targets $134 Billion Valuation in New Funding Round | PYMNTS.com

Published

on

Data analytics/artificial intelligence (AI) firm Databricks is reportedly raising $4 billion in a new funding round.

This Series L round would value the company at $134 billion, up 34% from its last session of funding during the summer, the Wall Street Journal (WSJ) reported Tuesday (Dec. 16).

Ali Ghodsi, Databricks’ co-founder and CEO, told the WSJ the company plans to use the new funding to invest in its core data-analytics products and AI software, while also letting its workers engage in secondary share sales.

The company, among the most valuable private firms in Silicon Valley, also plans to hire around 600 fresh college graduates in 2026, the CEO added, in addition to adding thousands of new jobs worldwide in Asia, Latin America and Europe. It also plans to hire AI researchers, who are typically paid top salaries, the WSJ added.

The report noted that Databricks has benefited from the AI boom, which relies partially on private corporate data to customize AI models. Databricks told the WSJ that its data-warehousing product, which can serve as an underlying data platform for AI services, surpassed a $1 billion revenue run rate at the end of October.

This year has seen Databricks ink deals with OpenAI and Anthropic to help sell AI services to business customers. Each of these partnerships are designed to push clients to develop AI agents, or independent bots that can carry out tasks on behalf of humans.

Advertisement: Scroll to Continue

The company’s new funding round comes three months after Databricks’ Series K round, which valued it more than $100 billion, up from $62 billion at the start of the year.

In other AI news, PYMNTS wrote earlier this week about The General Intelligence Company of New York, a start up developing agent-based systems designed to take over large portions of company operations.

“The company’s name deliberately evokes Gilded Age ambition, and founder Andrew Pignanelli told PYMNTS that the reference was intentional,” that report said. “He said he views AI as foundational infrastructure for the next era of company-building, much as railroads and industrial capital reshaped the United States economy more than a century ago.”

The company started by working backward from “the one-person billion-dollar business,” as Pignanelli termed it.

“We started at the end, the actual one-person billion-dollar company, and worked our way back and we were like, ‘What can we do today?’” he said.

Continue Reading

Fintech

Apple App Store Fees Face Pressure From EU Developers | PYMNTS.com

Published

on

A collection of app developers and consumer groups want Europe to enforce laws against Apple.

The Coalition of App Fairness (CAF) on Monday (Dec. 15) issued an open letter to the European Commission (EC) accusing the tech giant of “persistent” non-compliance with Europe’s Digital Markets Act (DMA).

The letter follows findings from the EC that Apple had violated the DMA by keeping developers from directing users to alternative payment methods, fining the tech giant $588 million.

Apple in turn revised its terms for its app store to impose fees that ranged from from 13% for smaller businesses to up to 20% for App Store purchases. However, the CAF says Apple has not addressed what it calls a core issue: the company’s fees are preventing fair competition.

“The law says that gatekeepers like Apple must allow developers to offer and conduct transactions outside of the App Store free of charge,” the letter said. “However, Apple is now charging developers commission, fees of up to 20% for such transactions. This is a blatant disregard for the law with the potential to vanquish years of meaningful work by the Commission.”

The CAF also notes that Apple plans to introduce new terms and conditions for the App Store next month, and says it suspects the new terms will include fees that violate the DMA.

Advertisement: Scroll to Continue

“Apple cannot be permitted to exploit its gatekeeper position by holding the entire industry hostage,” the letter added.

PYMNTS has contacted Apple for comment but has not yet gotten a reply. The company had in September called on the commission to rethink the DMA, which was created to prevent market abuse by tech giants doing business in Europe.

“Over that time, it’s become clear that the DMA is leading to a worse experience for Apple users in the EU,” Apple wrote in a blog post. “It’s exposing them to new risks, and disrupting the simple, seamless way their Apple products work together. And as new technologies come out, our European users’ Apple products will only fall further behind.”

In its blog post, Apple argued the DMA requirements for allowing other app marketplaces and alternative payment systems don’t take into account the privacy and security standards of the App Store, putting customers at risk for being overcharged or scammed.

“The DMA also lets other companies request access to user data and core technologies of Apple products,” the company wrote. “Apple is required to meet almost every request, even if they create serious risks for our users.”

Continue Reading

Trending