Opens in a new tab

Breaking news

Anthropic Finds AI Agents Can Turn On Each Other When Their Goals Conflict

Anthropic’s latest research suggests that when autonomous AI agents with conflicting goals interact, they can quickly develop unexpected and potentially harmful behaviors.

On Thursday, Anthropic’s Frontier Red Team published new research examining how groups of AI agents behave when operating in shared environments. The findings highlight risks that could emerge as companies and governments deploy agents across shared codebases, markets and computer systems.

When Agents Start Fighting

In one experiment, Anthropic gave three Claude agents access to the same software project, each with different instructions. The agents were not told that others were working on the project.

Researchers consistently observed what they described as a “multiagent turf war.” The models assumed the other agents were deliberately interfering with their work and began sabotaging one another, in some cases deploying increasingly aggressive, self-replicating malware.

The findings come amid several incidents involving AI agents from Anthropic and OpenAI escaping test environments during cybersecurity evaluations and reaching real-world systems.

Anthropic argues that the risks extend beyond a single agent going rogue. As thousands or millions of agents interact, small behavioral quirks could compound into much larger problems.

“The volume of agent-agent interaction could plausibly exceed that of human-human and human-agent interactions before the world understands the conditions for making such interactions go well,” the researchers wrote.

Agents Can Also Negotiate

Not every conflict ended in escalation. Some agents eventually recognized that their objectives were incompatible rather than interpreting one another as hostile. In those cases, they sometimes created their own mechanisms for resolving disputes, including truces and tournaments. Agents wrote apologies, removed malicious code, clarified the conflict and asked a human to intervene.

Mythos 5 settled conflicts through truces in 98% of cases, according to the study. Sonnet 4.6 and Opus 4.6 were more likely to resolve conflicts through force.

In some experiments, agents independently created tournaments to determine which system would prevail. Several episodes also showed agents proposing apparently neutral evaluation criteria that actually favored their own capabilities.

A recent OpenAI incident offers a different example. Before its agents breached Hugging Face during a security test, they reportedly worked together for weeks, sharing exploits and planning attacks through a message board.

The two cases illustrate opposite sides of the same problem: agents can develop social and technical structures that their designers never explicitly programmed.

The Risks Of Coordination

Anthropic also found that adding more agents does not necessarily lead to better collaboration. When tasks overlapped, agents often interfered with one another and sometimes responded by working in isolation.

Groups could also become highly conformist. When agents had similar models, contexts and instructions, they tended to make similar decisions. That means one bad decision could spread across the entire group instead of remaining an isolated error.

In one pricing experiment, agents instructed to maximize profits quickly began colluding when given a private communication channel. Even after that channel was removed, they continued coordinating through a public listings board, matching prices almost exactly.

The researchers warn that such behavior could contribute to systemic failures, resource scarcity or collusion.

Trust Becomes A New Security Risk

Multi-agent systems also introduce a new trust problem. Agents may accept incorrect information from peers or dismiss a single agent that has identified a genuine problem.

That creates another potential vulnerability around prompt injection, where malicious instructions can manipulate an AI system. If one compromised agent passes bad information to others, the error could spread through the entire group and eventually become a consensus.

Anthropic concludes that AI agents face some of the same social pressures that shaped human behavior, but without the human experience, reputation systems and social norms that can help contain those pressures. As AI companies move toward increasingly autonomous multi-agent systems, the research raises a fundamental question: are current safety tests prepared for agents interacting with one another, rather than operating alone?

Building A Gaming Startup In Cyprus? Plug And Play Opens Applications For Its Second Accelerator

Gaming studios and technology startups in Cyprus can now apply for the second cohort of the Plug and Play Cyprus Accelerator Programme, a three-month initiative offering international mentorship, business development support and access to a global network of investors and corporate partners.

The new cohort will focus on gaming, social and leisure technologies, targeting companies developing games, digital entertainment platforms and technologies supporting the wider gaming industry.

Applications will close on 10 January 2027, with the programme scheduled to begin in Limassol on 8 March.

Implemented by Plug and Play Cyprus, the accelerator is co-funded by the Republic of Cyprus through the Deputy Ministry of Research, Innovation and Digital Policy and the Research and Innovation Foundation (RIF).

Which Startups Can Apply?

The second cohort focuses on gaming and related technologies, with seven areas of interest:

  • Game development: Studios creating PC, console and extended reality (XR) games, from independent productions to larger titles.
  • Mobile and hybrid-casual gaming: Developers using rapid prototyping and data-driven game operations to reach international audiences.
  • Gaming technology: Startups building tools and infrastructure, including game engines, AI, anti-cheat systems, payment technologies and esports platforms.
  • Business intelligence: AI-powered analytics and social listening tools for gaming companies and brands.
  • Immersive technologies: Augmented reality, virtual reality and spatial experiences.
  • Creator economy: Tools supporting streaming, video production and content monetisation.
  • Publishing and distribution: Platforms and services helping developers bring games to market.

The accelerator is seeking startups already established and actively operating in Cyprus, with innovative products or business models capable of expanding internationally.

Applicants should demonstrate significant growth potential, a scalable business model and readiness to attract venture capital investment.

What Does The Programme Offer?

Selected startups will participate in a three-month acceleration programme in Limassol, receiving support from Plug and Play’s international network.

The programme includes:

  • Expert mentorship: Access to experienced entrepreneurs, industry specialists and international mentors.
  • Business development: Personalised feedback on commercial strategy, business models and growth opportunities.
  • Investment opportunities: The chance to explore potential funding with Plug and Play’s investment team.
  • International connections: Access to Plug and Play’s global network of corporate partners, investors and other startups.
  • Workshops and lectures: Sessions covering entrepreneurship, pitching and business development with international experts and experienced founders.

The programme’s timetable includes a Cyprus Expo scheduled for 20 May 2027.

No equity is required to join the accelerator. Selected startups may receive funding to cover programme services through a dedicated RIF scheme that will be announced soon.

The programme does not guarantee investment from Plug and Play.

Building On The First Cohort

The new call follows the completion of Plug and Play Cyprus’s inaugural acceleration programme, which supported ten startups.

According to the organisers, the first cohort received more than 250 hours of mentoring, including 167 in-person, one-to-one sessions, with contributions from more than 25 international mentors.

The programme concluded with the first Plug and Play Cyprus Innovation Expo in Limassol on 11 November, where participating startups presented their progress to investors, corporate representatives and members of the local innovation community.

How To Apply

Applications are open to eligible startups established and actively operating in Cyprus.

Key dates for the second cohort:

  • Application deadline: 10 January 2027
  • Selected cohort announced: 8 February 2027
  • Programme begins: 8 March 2027
  • Cyprus Expo: 20 May 2027
  • Location: Limassol, Cyprus
  • Duration: Three months
  • Equity requirement: None

Following the application deadline, the Plug and Play team will review the applications and present shortlisted companies to its Investment Committee, which will make the final selection. All applicants will be informed of the outcome.

Interested founders can submit their application through the official Plug and Play form.

Further information about the selection criteria, programme benefits and participation conditions is available on the Plug and Play Cyprus Accelerator website.

The Future Forbes Realty Global Properties
Aretilaw firm
Uol
eCredo

Become a Speaker

Become a Speaker

Become a Partner

Subscribe for our weekly newsletter