For several months, major AI companies have implemented specially approved programs and stringent safeguards to restrict the use of their models by malicious actors. However, these restrictions are now impairing the efforts of legitimate network defenders as well as offensive cybersecurity researchers. 

In June, the U.S. government imposed export control measures on Anthropic’s widely discussed AI models, Mythos and Fable. This action was partly triggered by a report suggesting that it might be feasible to circumvent the models’ safeguards designed to stop users from employing them to create and carry out harmful cyberattacks.

Regardless of whether the situation was genuinely influenced by concerns regarding a jailbreak, it’s evident that Anthropic has frequently promoted Mythos as a doomsday cyber tool that can only be entrusted to thoroughly vetted users, and even then with rigorous safeguards enforced. (The export restrictions on Fable 5 and Mythos 5 have since been rescinded. Fable 5 returned to public access on July 1; Mythos 5 has been reintegrated solely to screened U.S. organizations as part of the government’s evaluation process.)

This type of gatekeeping is not exclusive to Mythos. Both Anthropic with its other models and OpenAI offer cybersecurity researchers avenues to apply for vetting, and — if approved — gain access to models with reduced cybersecurity limitations: OpenAI’s Trusted Access for Cyber initiative and Anthropic’s Cyber Verification Program. 

These safeguards have faced significant criticism, especially from researchers tasked with uncovering unknown vulnerabilities within systems and formulating ways to exploit them before malicious actors do.

In a recent appearance on a cybersecurity podcast, Mark Dowd, a renowned security researcher, expressed that “it’s not really comfortable to me that these random large companies are making arbitrary decisions about what is safe in security and what’s not.”

Dowd has spent years identifying and selling “zero-days” — previously unidentified software flaws and the exploits that leverage them — to Western governments instead of reporting them to the software developers for patching. Governments pay a premium for vulnerabilities because they remain unaddressed, beneficial for intelligence activities.

Dowd acknowledged that his work might color his perspective, but he is not the only one. Multiple individuals working in offensive cybersecurity — who actively probe systems for weaknesses — described to TechCrunch their use of AI tools and how they navigate their guardrails. 

Chris Anley, the chief scientist at security consulting powerhouse NCC Group, mentioned that prompting an AI model to attempt to exploit a bug is crucial in verifying that it’s a genuine vulnerability and worth addressing. However, if a guardrail leads the model to outright refuse to respond, it adversely affects defenders, he stated.

“This is where the entire offensive versus defensive and guardrails aspect comes in, because ‘fix this code’ as a prompt serves as both a vital mechanism for defense and a blueprint for identifying critical vulnerabilities within the code base,” Anley explained. “Thus, the same tool functions as both an offensive and a defensive tool, and the two cannot actually be separated.”

“It’s ‘like a hammer,’” he continued. “You cannot construct a house without a hammer. It is undoubtedly a tool but irreducibly also a weapon.”

When he and his team encounter such hurdles, they sometimes revert to open-source AI models that come with no safeguards whatsoever.

Paolo Stagno, the chief technology officer at Crowdfense, a well-known entity that develops, procures, and sells undiscovered vulnerabilities to government bodies, shared Dowd’s sentiments, stating that AI companies “essentially treat clients like children who require supervision” with their vetted programs and guardrails. 

Stagno stated that he and his team do utilize frontier models — but strictly for reverse engineering. They refrain from employing AI to assist in identifying vulnerabilities or constructing exploits, as integrating that work into a cloud-based model poses risks of exposing sensitive vulnerability information or having it absorbed into future training sessions. For that purpose, he stated they resort to locally run open-source models, avoiding data sharing outside the model. 

Giuseppe Cali, a security researcher specializing in zero-days and exploit development, claimed that guardrails do not obstruct his work. This is because he does not use AI for offensive tasks; rather, he employs it for initial reverse engineering, to comprehend the code he’s analyzing, and to create supporting tools. For that, he emphasized that AI tools can expedite the process and allow him to concentrate on discovering vulnerabilities. 

“I still want to control the actual bug discovery and weaponization myself, and that wouldn’t change if all guardrails were removed tomorrow,” Cali remarked. “I am possessive about my bugs, and I enjoy this game too much to allow models to play it for me.”

One researcher at a smartphone-component manufacturer, who requested anonymity due to lack of authorization to speak to the media, indicated that his employer is not part of Anthropic’s CVP program, leaving its tools nearly ineffective for discovering vulnerabilities due to overly strict guardrails.

“If it catches wind we’re doing anything security related, it just stops and isn’t usable,” the person remarked. 

Chris Thompson — chief executive of cybersecurity firm RemoteThreat and founder of Offensive AI Con, an event focused on offensive security and AI — stated that in his experience with frontier AI models, the guardrails can be erratic and vary in performance daily. This inconsistency holds true even within the more lenient frameworks of Anthropic’s and OpenAI’s vetted programs. 

“I think the practical effect is that you spend a lot of time negotiating with the model instead of concentrating on the fundamental security program,” Thompson remarked. “Rather than analyzing a vulnerability and reasoning through the exploitability, you are attempting to uncover why you are encountering inconsistent results or why models are over-sanitizing the output.” 

Consequently, researchers are driven towards or rely on Chinese open-source models like GLM — models that can be freely downloaded and executed locally without any vetting or usage constraints — noted Thompson.

“You have these responsible researchers that are being pushed away from U.S.-governed systems to foreign-owned systems,” he remarked. “I believe it’s more damaging than beneficial to have these guardrails in place.”

Instead of imposing further restrictions, Thompson urged the AI frontier labs to expand their programs, provide responsible access, and hold those who misuse their tools accountable. Otherwise, he contended, defenders will lose the AI race.

“There’s this big storm approaching. There’s this significant wave of attacks that are going to occur at unprecedented speed and scale,” Thompson warned. “But the same security consulting companies and legitimate researchers striving to make a difference are currently being stifled.”

TechCrunch Startup Battlefield is heading to Australia — and we’ve teamed up with Stripe to discover the nation’s most thrilling early-stage startups. On August 19, eight Australian startups will present live at Stripe Tour Sydney in front of investors, media, and the tech community. Three of them will be awarded prizes. One will secure an automatic entry to TechCrunch Disrupt 2026, occurring in San Francisco from October 13-15.

What’s the key to identifying the next standout founder? The judges.

Startup Battlefield has a proven history of success. The competition has launched more than 1,700 companies worldwide, including Dropbox, Cloudflare, Discord, and Trello. Alumni have collectively raised $32 billion and achieved over 250 exits. During the last Startup Battlefield in Sydney, HealthMatch emerged as the winner — now valued at over $25 million with a million patients globally.

Such success isn’t coincidental. It stems from having the right individuals present, those who can identify potential, pose challenging questions, and create opportunities for founders. This year, we’ve gathered three judges representing the pinnacle of Australia’s startup landscape and global fintech innovation.

Meet your judges

Jasmine Liew, Head of Startup & Investor Partnerships for APAC, Stripe

Stripe not only sponsors Startup Battlefield — the company is also intricately woven into the startup landscape. Jasmine Liew possesses institutional insight into what it entails to scale on a global level and the payment infrastructure that underpins contemporary commerce. When a founder presents before her, they are pitching to someone who comprehends not just the vision, but also the intricacies of transforming that vision into a profitable venture.

Brendan Hill, angel investor and venture partner, Ten13

Brendan Hill embodies the investor segment that actively provides funding. As a venture partner and engaged angel investor, he not only assesses ideas — he’s also considering whether he would invest his own capital in a team. That perspective is crucial when founders seek to secure their subsequent round.

Jarron Aizen, founder and CEO, Hapana

Jarron Aizen is a founder himself, bringing his understanding of the experience these eight pitchers will face. He’s traversed the journey from concept to company, and that insight is incredibly valuable. When he poses a question from the judge’s panel, it is backed by the triumphs and challenges of developing a business in the Australian landscape.

What’s at stake

The grand prize champion will receive $15,000 in Stripe fee credits — tangible funds that can expedite growth — along with automatic entry into Startup Battlefield 200 at TechCrunch Disrupt 2026. This is not merely a reward. It’s a launchpad. The second place winner will get $5,000 in Stripe credits, and the third place winner will receive $2,000 in Stripe credits. More significantly, all finalists will have the opportunity to interact with these judges, gain media exposure, and connect with the Stripe Tour Sydney audience comprised of founders, operators, and investors. 

The true spectacle occurs on August 19

If you’re an investor, a founder not ready to pitch, or simply someone keen to witness the evolution of Australian startups, Stripe Tour Sydney on August 19 is the event to attend. Register now.

The stage is ready. The judges await. The question remains: Which startup will shine next?

Hosted and MC’d by Isabelle Johannessen, director of Startup Battlefield.

Chip manufacturer AMD is setting its sights on rival Nvidia with its newest hardware introduction: a rack-scale solution crafted to meet the computing demands of the globe’s largest AI laboratories.

At the company’s packed Advancing AI conference in San Francisco on Thursday, AMD Chair and CEO Dr. Lisa Su showcased the new AI rack system named Helios — along with its increasing roster of clients, which includes Microsoft — as the firm gears up for its release later this year. Su also promoted the firm’s latest chips designed to satisfy the high demands of the AI sector.

Rack systems merge multiple processors into one robust unit. They are specifically designed for data centers, where they facilitate the training and operation of AI models and other computation-heavy tasks.

Su described Helios as the industry’s “top-performing AI rack,” noting that it was “engineered to train and execute the most demanding frontier models globally at an unprecedented scale.” The system is set to be implemented by prominent AI enterprises at gigawatt-scale, according to the company.

Historically, Nvidia has led this field with its Vera Rubin and Grace Blackwell rack-scale systems. AMD is clearly aiming to get a piece of the pie. Helios’ performance indicators seem to offer it a tangible opportunity, reportedly surpassing Vera Rubin in various metrics, as noted by The Register.

Helios, unveiled in 2025 and presented live in January at CES 2026, already counts some notable clients among its base, such as OpenAI, Meta, Oracle, Anthropic, and Microsoft, all of which intend to utilize the system. Microsoft CEO Satya Nadella mentioned on Monday that the company plans to enhance its Azure infrastructure with Helios. Additionally, Anthropic and AMD declared a strategic alliance on Wednesday to deploy up to two gigawatts of GPUs through the new rack system.

On Thursday, AMD also unveiled its Venice-X CPU, engineered for data centers and capable of managing high-computing workloads. The Venice-X is projected for launch in 2027.

During her speech, Su reflected on the chip industry’s path, asserting that by 2030, chips supporting AI will constitute a significant portion of the overall computing market. This transformation is propelled by a “step change in compute demand,” largely fueled by the emergence of agentic AI, she stated.

“When you instruct the agent to perform a task, it goes through numerous steps, reasoning, calling tools, accessing data, and repeating the process until it resolves the issue; thus, a multitude of GPUs is required to manage all of that,” the executive explained.

“We anticipate that by 2030, the AI accelerator market will reach approximately $1.4 trillion,” Su stated. “This indicates that by decade’s end, the AI accelerator market will come close to matching the total size of the current semiconductor market.”

“We do foresee that GPUs will represent the bulk of that market since the algorithms are still in their early stages, and we’re observing continual shifts in workloads, which favors programmability within the complete silicon ecosystem,” she added.

Patreon is reducing its workforce by 20%, which equates to 93 employees, as announced by CEO Jack Conte to staff on Thursday. In a memorandum that the company made public, Conte expressed that while Patreon’s primary operations remain strong, the platform must adapt to market shifts and modify its cost framework to maintain stability, hence the need for these “painful” but essential layoffs.

Conte stated that “AI has drastically altered the technology sector,” and the rate of change has “never been more rapid.” Nevertheless, he emphasized that Patreon is not implementing these layoffs to replace employees with AI.

“To clarify the role of AI in today’s decision: we are not enacting the above changes because we think AI can replace human workers,” he noted. “The more we have embraced these new tools, the more evident it has become that they cannot match the creativity, judgment, attention to detail, or craftsmanship that our team possesses abundantly, nor do they satisfy the human connection that we all value deeply. That’s my personal viewpoint, but more significantly, it forms the basis of Patreon’s strategy: our product vision and business are fundamentally based on the worth of human creativity and connection.”

He added, “AI has indeed transformed the tech landscape, including our modes of work, product development, communication, and beyond. This does influence our operations and organizational structure.”

Alongside the reduction in workforce, Conte stated that Patreon is also reorganizing its operational structure, “flattening” its hierarchy and realigning teams with its primary objectives. Those affected will receive a minimum of 16 weeks of severance pay, plus an extra week for each year of service, healthcare coverage until year-end, and a $1,500 allowance to replace their last company-issued laptop.

Last week, Patreon revealed its collaboration with internet infrastructure provider Cloudflare to proactively prevent AI bots from accessing creators’ work without consent to train their AI models. The company indicated that it needed to bolster its defenses in this area as AI scraping methods have evolved. This action comes as online publishers and creators face challenges from AI companies utilizing their work for training AI models.

The most recent layoffs at Patreon mark the largest reduction since a 17% staff cut in 2022. In that previous round, Patreon also shuttered its offices in Berlin and Dublin.

Corgi, a startup specializing in insurance technology, data room solutions, and coffee shops, is reportedly in the process of raising another funding round that closely follows its previous fundraising and would effectively double its valuation, according to sources cited by Forbes.

This round is anticipated to be a second extension of its Series B funding and has already been finalized. Corgi disclosed its last funding round, the B1 round, which amounted to $106 million at a valuation of $2.6 billion, just eight weeks ago at the end of May.

Amidst the current surge in AI funding, numerous startups are securing consecutive rounds at escalating valuations — yet Corgi has managed to stand out even in that crowded field.

The Y Combinator graduate (summer 2024) secured a $108 million Series A in January at an undisclosed valuation. (PitchBook approximates it at $630 million post-money.) Four months later, it raised its Series B with $160 million at a valuation of $1.3 billion. Just three weeks post that, it revealed a B1 round where the same investors contributed $106 million at a $2.6 billion valuation.

Now, sources inform Forbes of a B2 round emerging eight weeks later. However, Forbes did not disclose the amount that Corgi raised, and the company opted not to comment on the potential funding.

Corgi is supported by TCV and Kindred Ventures. Kanyi Maqubela from Kindred highlighted the startup’s progress to TechCrunch as a rationale for the recent valuation increase, attributing it to the company’s revenue growth.

When Corgi disclosed its Series A seven months back, the founders indicated that the firm had already achieved a $40 million annualized revenue rate. Sources have informed Forbes that it’s now positioned to elevate that run rate to $450 million by the year’s end.

Corgi offers AI-driven insurance solutions, leveraging AI to provide rapid quotes and expedite claims payments. It provides various forms of liability insurance for startups, including general liability, technology-related incident coverage, and employment liability, alongside business renters’ and auto insurance.

While insurance is typically a capital-intensive industry, Corgi faces additional challenges due to its use of a Risk Retention Group (RRG) structure. This method allows individuals in similar sectors or facing like liabilities to pool their resources for collective self-insurance.

As stated on Corgi’s website, RRGs are not governed by all the same state regulations that traditional, rated, underwritten insurance carriers are subject to. According to a spokesperson, Corgi has employed various structures for different insurance types; some policies may utilize state-regulated carriers as an example.

However, in an RRG insurance framework, claims are disbursed from the pooled funds, meaning a significant claim can diminish the resources available for settling other claims. RRGs lack the backing of state guaranty funds; if the pool fails to cover claims, the members absorb the losses. Large claims can even lead to the bankruptcy of the RRG.

Therefore, it’s perhaps understandable that Corgi seeks to enhance its financial reserves.

Despite the expansion of its insurance offerings, the startup has also diversified in other aspects. Corgi has introduced data room software, successfully navigating a recent controversy regarding the software’s coding.

Additionally, the startup operates two 24/7 coffee shops with charming and sometimes sponsored drink names like “Brexspresso.” Located in San Francisco and Atlanta, Corgi plans to open five more locations soon, including several in New York and one in London. Establishing physical cafe locations also entails significant funding.

In the meantime, Corgi has developed a reputation within Silicon Valley for its demanding corporate environment, following remarks from founder-CEO Nico Laqua regarding his expectations for employees to work seven days a week.