Anthropic Reassesses Safety Commitments Amidst Competitive Pressure
Anthropic, the artificial intelligence firm renowned for the Claude chatbot and dedicated to the development of safe technologies, has recently indicated a retreat from its previously stringent safety commitments in a bid to maintain its competitive edge.
On Tuesday, the company unveiled an alteration to its responsible-scaling policy, which comprises self-imposed guidelines designed to mitigate the risks associated with the proliferation of potentially harmful AI technologies, including the threat of large-scale cyber incursions.
The revised guidelines stipulate that while Anthropic will continue to demand a “strong argument” that catastrophic risks are sufficiently managed during AI development, they now signal a willingness to proceed with development activities as long as the company perceives a competitive advantage over rivals.
This strategic pivot is reportedly in response to a shifting landscape in U.S. regulatory attitudes, where the prioritisation of economic potential appears to supersede concerns regarding AI safety.
“Despite the rapid evolution of AI capabilities in recent years, governmental measures addressing AI safety have lagged considerably,” the company noted in a recent blog entry.
Moreover, Anthropic pointed out that the policy climate now favours competitive AI growth, while substantive safety dialogues have yet to manifest at the federal level.
This modification in Anthropic’s safety protocols coincides with tensions arising from the Pentagon’s intention to withdraw contracts unless its technology is permitted for all lawful military applications — a development Anthropic insists is unrelated.
Historically, Anthropic has marketed itself as a proponent of safety-first principles.
Founded in 2021 by former OpenAI employees disenchanted with that company’s prioritisation of development over safety, CEO Dario Amodei has vocalised apprehensions regarding AI’s potential for adverse societal impacts, including catastrophic scenarios, while affirming that safety remains an overarching priority for the organisation, as stated in a December interview with Fortune.
Anthropic CEO and co-founder Dario Amodei, seen in January at the World Economic Forum in Davos, Switzerland, emphasised in a December interview that safety continues to be the company’s ‘highest-level focus.’
In its blog post, Anthropic underscored that safety protocols are intended to evolve, claiming this latest modification enhances “transparency and accountability” through new commitments to the regular publication of safety reports and goals.
However, Heidy Khlaaf, chief AI scientist at the independent think tank AI Now Institute, criticised the company’s reassessment, arguing that despite its safety-centric branding, Anthropic has historically underperformed in its efforts to avert human harm.
Cross Country Checkup is asking: How are you using AI? Is AI making our lives better or worse? Leave your comment here, and we may read it or call you back for Sunday’s show.
Khlaaf pointed out that, from its inception, Anthropic has disproportionately emphasised the potential for future catastrophic repercussions while neglecting the immediate hazards inherent in current AI applications, such as typical errors generated by chatbots.
In instances past, the Claude chatbot has been exploited in fraudulent schemes and malware creation, and it has recently been leveraged to unlawfully obtain sensitive Mexican government data, as reported by cybersecurity analysts.
Khlaaf contends that the company is now abandoning the “veneer of safety” it previously employed as a marketing tactic, recognising that such an approach is no longer advantageous.
“This represents a strategic pivot aimed at manifesting their readiness for commerce,” Khlaaf asserted.
Canada launches AI watchdog to oversee the technology’s safe development and use
This announcement arrives amidst an extraordinarily competitive environment among premier AI entities, including Anthropic, OpenAI, and Google, all of which aim to integrate their technologies into business and government frameworks.
The U.S. administration, under President Trump, has also indicated an unwavering commitment to AI advancement, threatening to withhold federal funding from states that enact regulations seen as detrimental to U.S. competitiveness in the sector.
Teresa Scassa, Canada Research Chair in Information Law and Policy at the University of Ottawa, articulated that this lack of regulatory guidance from the U.S. could hinder companies focused on safety since adhering to such protocols may render them uncompetitive.
This predicament poses challenges for Canada as well, Scassa noted, suggesting that stringent regulations could stifle domestic AI innovation or spur Canadian firms to relocate to the U.S., where regulatory frameworks are more accommodating.
“It’s evident that Canada cannot afford to lose its competitive standing in this arena,” Scassa explained. “This dynamic influences the trajectory of AI regulation here.”
Since the collapse of the Artificial Intelligence and Data Act in 2025, there has been a conspicuous absence of any comprehensive AI regulatory efforts in Canada, mirroring the regulatory void present in the United States.
Safety Alteration Not Linked to Pentagon Dispute, Company Asserts
The modification to Anthropic’s safety policies emerges alongside pressures from the Pentagon.
In July, Anthropic entered into a contract with the U.S. Department of Defence, valued at up to $200 million, permitting the government to utilize its technology for military objectives, albeit strictly adhering to the company’s usage guidelines, which govern the permissible applications of its products, including the Claude chatbot.
Front Burner21:47Will AI agents take over the workplace?
These guidelines explicitly prohibit any user, including federal authorities, from employing Anthropic’s AI tools for purposes such as weapon design or development.
Nonetheless, reports indicate that U.S. Defence Secretary Pete Hegseth presented CEO Amodei with an ultimatum during a Tuesday meeting — mandating that the company permit military use of its AI technologies for all legal purposes by Friday, or face the revocation of its government contracts and designation as a supply chain risk.
Front Burner21:47Will AI agents take over the workplace?
In its ongoing negotiations with the government, Anthropic has affirmed that it will not permit its technology to be utilised in autonomous weaponry — systems that allow AI to independently select and engage targets — or for mass surveillance.
However, Pentagon representatives have conveyed to the media that the contention does not pertain to the potential applications of AI in autonomous weaponry or mass surveillance, asserting that the government has consistently complied with legal frameworks.
Anthropic maintains that the revision of its responsible scaling policy and the Pentagon’s demands are unrelated. According to the company, Hegseth’s concerns are more closely aligned with its usage policy than its scaling policy.
In light of the impending deadline, Amodei remarked in a blog post that Anthropic would resist the administration’s requests, reaffirming the company’s opposition to the use of its technology in domestic surveillance and autonomous weapon systems.
Amodei expressed hope that the Pentagon might reassess its stance but indicated that the company is prepared to facilitate a smooth transition to alternative providers should the Pentagon opt to terminate the contract.
“Our primary preference is to continue serving the department and our military personnel — with our stipulated safeguards intact,” Amodei articulated.
Source link: Cbc.ca.
